{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 301 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "114", "answer_2": "109", "answer_3": "123", "answer_0": "126"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "14", "answer_0": "28", "answer_1": "16", "answer_3": "14"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 5 people. A shopper bought a 25 gram box and a 250 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "15", "answer_0": "27", "answer_1": "17", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.125", "p1": "0.375", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $394. A fee of $183 is charged, then a deposit of $81 arrives. A second fee of $91 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "228", "answer_3": "230", "answer_0": "201", "answer_1": "181"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Thursday, November 05, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.600", "p1": "0.333", "p2": "0.333", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 7 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed June 06, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 13", "date_0": "June 12", "date_2": "June 10", "date_1": "June 11"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p0": "0.500", "p2": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $407 per year. It increases by 10% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "452.7", "answer_2": "366.3", "answer_0": "436.7", "answer_1": "447.7"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_3": "35", "answer_2": "25", "answer_1": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $714. A fee of $161 is charged, then a deposit of $60 arrives. A second fee of $80 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "559", "answer_0": "533", "answer_2": "515", "answer_1": "519"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = $342; priority = low.\n\nClaim to verify: \"the owner is $342\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.583", "p2": "0.417", "p1": "0.417", "p3": "0.545"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 269 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "187", "answer_0": "200", "answer_3": "193", "answer_1": "185"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "32", "answer_0": "80", "answer_1": "52", "answer_2": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.385", "p2": "0.571", "p1": "0.571", "p0": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 968:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 382 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_3": "65", "answer_2": "64", "answer_1": "74"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 3 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 12:15 to 17:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "5h30", "dur_0": "5h15", "dur_3": "4h30", "dur_1": "5h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = $147; priority = high.\n\nClaim to verify: \"the status is $147\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p1": "0.667", "p0": "0.333", "p2": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "54", "age_2": "53", "age_0": "52", "age_1": "51"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 424:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = 2026-03-12; priority = high.\n\nClaim to verify: \"the status is 2026-03-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = 2026-06-18; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $864 per year. It increases by 10% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "777.6", "answer_0": "930.4", "answer_3": "912.4", "answer_1": "950.4"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "35", "answer_3": "25", "answer_2": "10", "answer_1": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "19", "answer_3": "21", "answer_0": "33", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = $75; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $30k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed April 16, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 24", "date_3": "April 24", "date_1": "April 22", "date_2": "April 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 296 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "290", "answer_0": "288", "answer_2": "303", "answer_1": "304"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.417", "p2": "0.250", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 187 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "97", "answer_0": "105", "answer_3": "95", "answer_2": "116"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = 2026-09-12; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed September 17, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 23", "date_2": "September 21", "date_3": "September 23", "date_1": "September 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed June 20, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 25", "date_3": "June 27", "date_2": "June 28", "date_0": "June 26"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $569 per year. It increases by 5% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "558.45", "answer_2": "540.55", "answer_1": "597.45", "answer_0": "573.45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = 2026-07-24; priority = high.\n\nClaim to verify: \"the status is 2026-07-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_1": "30", "answer_0": "48", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1977. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "48", "age_2": "50", "age_0": "49", "age_3": "51"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 7 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 312 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "140", "answer_2": "132", "answer_3": "132", "answer_1": "155"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed September 24, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 29", "date_2": "September 29", "date_3": "October 01", "date_0": "October 01"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = 2026-11-16; priority = low.\n\nClaim to verify: \"the amount is 2026-11-16\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 5 grams of salt per serving and serves 6 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 9 grams of salt per serving and serves 6 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed June 15, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 24", "date_1": "June 24", "date_3": "June 26", "date_0": "June 26"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $849 per year. It increases by 15% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "948.35", "answer_1": "976.35", "answer_3": "940.35", "answer_2": "721.65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Thursday, June 04, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = $55; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.444", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1990. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "35", "age_3": "34", "age_0": "36", "age_2": "37"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $27k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = 2026-03-26; priority = low.\n\nClaim to verify: \"the status is 2026-03-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = 2026-05-21; priority = low.\n\nClaim to verify: \"the status is 2026-05-21\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed December 16, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 20", "date_1": "December 20", "date_3": "December 22", "date_0": "December 22"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $311 per year. It increases by 10% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "342.1", "answer_0": "300.1", "answer_2": "279.9", "answer_3": "336.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed October 28, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 06", "date_2": "November 08", "date_1": "November 04", "date_3": "November 06"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = $949; priority = high.\n\nClaim to verify: \"the date is $949\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.583", "p0": "0.417", "p1": "0.583", "p3": "0.364"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = $225; priority = low.\n\nClaim to verify: \"the status is $225\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "8", "answer_2": "16", "answer_1": "18", "answer_0": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $34k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 20:45 to 22:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h15", "dur_1": "1h45", "dur_0": "1h00", "dur_3": "0h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, May 17, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1962. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "64", "age_1": "62", "age_3": "61", "age_0": "63"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1791 per year. It increases by 5% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1701.45", "answer_3": "1872.55", "answer_0": "1856.55", "answer_1": "1880.55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=153, unit_price=423, discount_pct=0, region=EU, sku_total=576 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "64566", "v2": "6471", "v0": "64719", "v1": "576"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p0": "0.800", "p3": "0.778", "p2": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_3": "6", "answer_0": "18", "answer_1": "14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $448 per year. It increases by 10% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "403.2", "answer_0": "464.8", "answer_1": "492.8", "answer_3": "501.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = $634; priority = low.\n\nClaim to verify: \"the status is $634\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed October 26, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 07", "date_1": "November 05", "date_0": "November 09", "date_2": "November 07"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_1": "20", "answer_3": "6", "answer_2": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1970. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "57", "age_3": "54", "age_1": "55", "age_0": "56"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = open; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed September 04, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 11", "date_2": "September 09", "date_1": "September 09", "date_3": "September 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed June 02, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 07", "date_1": "June 07", "date_0": "June 09", "date_3": "June 09"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.462", "p1": "0.500", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 935:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = closed; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 6 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, August 05, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "20", "answer_2": "8", "answer_1": "13", "answer_0": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1975. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "51", "age_0": "50", "age_3": "48", "age_1": "49"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.625", "p1": "0.333", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.083", "p1": "0.333", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1659 per year. It increases by 20% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1982.8", "answer_0": "1959.8", "answer_1": "1990.8", "answer_2": "1327.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $213. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1263 per year. It increases by 5% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1199.85", "answer_1": "1326.15", "answer_0": "1300.15", "answer_3": "1276.15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1972. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "55", "age_0": "54", "age_1": "53", "age_3": "52"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = $744; priority = low.\n\nClaim to verify: \"the owner is $744\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, June 26, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Wednesday, February 25, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Friday, September 11, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, February 08, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.222", "p3": "0.125", "p2": "0.778", "p1": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.167", "p2": "0.167", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.667", "p2": "0.500", "p1": "0.833"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 282 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "65", "answer_2": "31", "answer_1": "42", "answer_0": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $518. A fee of $127 is charged, then a deposit of $57 arrives. A second fee of $63 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "395", "answer_3": "356", "answer_0": "385", "answer_1": "378"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 11:45 to 19:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "7h15", "dur_2": "7h45", "dur_3": "6h45", "dur_1": "7h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_0": "77", "answer_2": "42", "answer_1": "47"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.333", "p1": "0.417", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 253:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed April 20, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 26", "date_2": "April 30", "date_0": "April 28", "date_3": "April 28"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "21", "answer_2": "42", "answer_1": "45", "answer_0": "63"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = $19; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 969:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 218:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $609. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $795 per year. It increases by 10% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "862.5", "answer_1": "874.5", "answer_3": "824.5", "answer_2": "715.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p1": "0.200", "p3": "0.050", "p0": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $436 per year. It increases by 15% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "488.4", "answer_0": "455.4", "answer_2": "370.6", "answer_1": "501.4"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $411. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $708. A fee of $199 is charged, then a deposit of $45 arrives. A second fee of $99 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "432", "answer_2": "429", "answer_1": "461", "answer_0": "455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p2": "0.625", "p0": "0.375", "p3": "0.286"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 8 grams of salt per serving and serves 10 people. A shopper bought a 80 gram box and a 800 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 336 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "221", "answer_0": "240", "answer_3": "245", "answer_2": "243"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=154, unit_price=819, discount_pct=0, region=EU, sku_total=973 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "973", "v0": "126126", "v3": "125972", "v2": "12612"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $422. A fee of $35 is charged, then a deposit of $79 arrives. A second fee of $17 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "454", "answer_3": "451", "answer_1": "478", "answer_0": "449"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed September 20, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 26", "date_0": "September 28", "date_3": "September 28", "date_2": "September 26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1985. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "37", "age_0": "39", "age_2": "40", "age_1": "38"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 342:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = $81; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 336 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "135", "answer_3": "126", "answer_0": "144", "answer_1": "132"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $25k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.143", "p1": "0.750", "p2": "0.750"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $700. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 903:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=703, unit_price=157, discount_pct=0, region=EU, sku_total=860 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "860", "v0": "110371", "v2": "11037", "v3": "111074"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $28k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = $417; priority = high.\n\nClaim to verify: \"the date is $417\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 731:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = $597; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 13:15 to 20:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "6h30", "dur_1": "7h30", "dur_0": "6h45", "dur_2": "7h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=521, unit_price=247, discount_pct=0, region=EU, sku_total=768 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "12868", "v0": "128687", "v3": "129208", "v1": "768"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p3": "0.050", "p1": "0.250", "p0": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 8 grams of salt per serving and serves 3 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.556", "p2": "0.444", "p1": "0.444", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $511 per year. It increases by 10% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "459.9", "answer_3": "530.1", "answer_0": "546.1", "answer_1": "562.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1006 per year. It increases by 5% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1022.3", "answer_2": "955.7", "answer_0": "1009.3", "answer_1": "1056.3"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p1": "0.250", "p0": "0.200", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 07:15 to 12:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "5h15", "dur_3": "4h30", "dur_1": "5h30", "dur_0": "5h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.400", "p0": "0.600", "p1": "0.400", "p3": "0.556"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 206:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 53 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "60", "answer_1": "89", "answer_3": "91", "answer_0": "72"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.333", "p3": "0.083", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $211. A fee of $125 is charged, then a deposit of $75 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "105", "answer_0": "99", "answer_3": "94", "answer_2": "94"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed May 06, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 17", "date_0": "May 19", "date_2": "May 17", "date_1": "May 15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "28", "answer_3": "24", "answer_0": "48", "answer_2": "24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_1": "21", "answer_3": "20", "answer_0": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.200", "p0": "0.150", "p2": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 626:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = $860; priority = low.\n\nClaim to verify: \"the amount is $860\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 10:00 to 16:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "6h15", "dur_2": "6h15", "dur_0": "5h45", "dur_3": "5h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 379:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, September 09, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 5 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 447:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $468 per year. It increases by 5% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "444.6", "answer_3": "491.4", "answer_0": "460.4", "answer_1": "491.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed October 20, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 31", "date_0": "October 29", "date_1": "October 27", "date_3": "October 29"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = $613; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Sunday, May 10, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $413 per year. It increases by 5% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "392.35", "answer_3": "430.65", "answer_0": "415.65", "answer_1": "433.65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 07:45 to 08:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h30", "dur_1": "0h30", "dur_0": "-1h45", "dur_2": "0h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "15", "answer_2": "12", "answer_3": "9", "answer_0": "21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=802, unit_price=941, discount_pct=0, region=EU, sku_total=1743 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "754682", "v1": "1743", "v2": "75468", "v3": "755484"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $3k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 406:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "40", "answer_2": "32", "answer_0": "72", "answer_1": "37"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, June 23, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=705, unit_price=325, discount_pct=0, region=EU, sku_total=1030 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "229830", "v2": "22912", "v0": "229125", "v1": "1030"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 08:45 to 16:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "7h45", "dur_2": "8h00", "dur_3": "7h00", "dur_1": "8h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 9 grams of salt per serving and serves 3 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 719:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, November 21, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 83 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "180", "answer_3": "198", "answer_2": "175", "answer_1": "165"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $817. A fee of $32 is charged, then a deposit of $32 arrives. A second fee of $16 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "784", "answer_3": "787", "answer_0": "801", "answer_2": "781"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 571:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = 2026-08-12; priority = high.\n\nClaim to verify: \"the owner is 2026-08-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.750", "p2": "0.250", "p1": "0.250", "p3": "0.714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1980. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "44", "age_0": "45", "age_2": "46", "age_3": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_0": "56", "answer_1": "21", "answer_3": "40"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.500", "p2": "0.167", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $337. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.429", "p2": "0.429", "p0": "0.571", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $85. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 357 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "131", "answer_2": "130", "answer_1": "93", "answer_0": "112"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.417", "p3": "0.083", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_3": "18", "answer_0": "36", "answer_1": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 270 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "74", "answer_3": "65", "answer_0": "63", "answer_2": "50"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed August 16, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 22", "date_2": "August 22", "date_3": "August 24", "date_0": "August 24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = 2026-01-08; priority = high.\n\nClaim to verify: \"the owner is 2026-01-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 148 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_2": "53", "answer_3": "63", "answer_1": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = 2026-08-20; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $289. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $27k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1057 per year. It increases by 15% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1215.55", "answer_0": "1196.55", "answer_3": "1226.55", "answer_2": "898.45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = $356; priority = high.\n\nClaim to verify: \"the date is $356\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed October 07, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 18", "date_3": "October 16", "date_0": "October 16", "date_1": "October 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 6 grams of salt per serving and serves 12 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $337. A fee of $178 is charged, then a deposit of $74 arrives. A second fee of $89 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "144", "answer_1": "173", "answer_2": "125", "answer_3": "141"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, March 04, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=287, unit_price=403, discount_pct=0, region=EU, sku_total=690 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "115948", "v1": "690", "v2": "11566", "v0": "115661"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 971:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1981. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "44", "age_2": "45", "age_3": "46", "age_1": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, July 14, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=716, unit_price=199, discount_pct=0, region=EU, sku_total=915 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "915", "v0": "142484", "v2": "14248", "v3": "143200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 494:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $298. A fee of $173 is charged, then a deposit of $64 arrives. A second fee of $86 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "94", "answer_1": "98", "answer_0": "103", "answer_3": "77"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 851:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=352, unit_price=913, discount_pct=0, region=EU, sku_total=1265 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "321728", "v1": "1265", "v2": "32137", "v0": "321376"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $234 per year. It increases by 15% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "198.9", "answer_1": "269.1", "answer_0": "240.1", "answer_3": "280.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.200", "p3": "0.050", "p2": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $158. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1971. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "52", "age_0": "53", "age_3": "51", "age_2": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, May 23, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = 2026-12-21; priority = high.\n\nClaim to verify: \"the owner is 2026-12-21\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1962. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "64", "age_3": "66", "age_2": "65", "age_1": "63"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1968. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_3": "60", "age_1": "57", "age_2": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = 2026-12-24; priority = low.\n\nClaim to verify: \"the amount is 2026-12-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "22", "answer_3": "48", "answer_2": "16", "answer_0": "64"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=194, unit_price=194, discount_pct=0, region=EU, sku_total=388 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "388", "v3": "37830", "v0": "37636", "v2": "3763"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = 2026-07-15; priority = low.\n\nClaim to verify: \"the owner is 2026-07-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_1": "16", "answer_0": "28", "answer_2": "14"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=772, unit_price=731, discount_pct=0, region=EU, sku_total=1503 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "56433", "v1": "1503", "v3": "565104", "v0": "564332"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1984. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "41", "age_2": "42", "age_1": "40", "age_3": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $574 per year. It increases by 10% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "628.4", "answer_0": "618.4", "answer_2": "516.6", "answer_1": "631.4"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $572. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1158 per year. It increases by 20% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "926.4", "answer_1": "1389.6", "answer_0": "1365.6", "answer_3": "1370.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Thursday, July 02, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1977. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "48", "age_2": "49", "age_3": "46", "age_1": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p3": "0.050", "p1": "0.200", "p2": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $9k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed December 17, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 21", "date_0": "December 23", "date_2": "December 25", "date_3": "December 23"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 540:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $900. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=917, unit_price=979, discount_pct=0, region=EU, sku_total=1896 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1896", "v2": "89774", "v0": "897743", "v3": "896826"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.100", "p1": "0.200", "p0": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.700", "p3": "0.222", "p0": "0.300", "p1": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed April 22, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 07", "date_3": "May 03", "date_1": "May 01", "date_0": "May 05"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.500", "p1": "0.667", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = 2026-04-07; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $851. A fee of $83 is charged, then a deposit of $38 arrives. A second fee of $41 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "780", "answer_0": "765", "answer_2": "786", "answer_1": "743"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 7 grams of salt per serving and serves 4 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Sunday, July 05, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $27k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.800", "p0": "0.200", "p2": "0.800", "p3": "0.111"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.615", "p3": "0.333", "p0": "0.385", "p2": "0.615"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=429, unit_price=310, discount_pct=0, region=EU, sku_total=739 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "133419", "v2": "13299", "v1": "739", "v0": "132990"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 3 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed June 14, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 21", "date_2": "June 21", "date_1": "June 19", "date_0": "June 19"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Friday, August 28, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, November 26, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1985. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "42", "age_1": "40", "age_3": "43", "age_0": "41"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = $914; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 14:45 to 18:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h15", "dur_3": "2h15", "dur_0": "3h00", "dur_1": "3h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $2k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.545", "p1": "0.455", "p2": "0.455", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 11 people. A shopper bought a 44 gram box and a 440 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1983. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "42", "age_3": "41", "age_0": "43", "age_2": "44"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = 2026-11-12; priority = high.\n\nClaim to verify: \"the status is 2026-11-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 18:45 to 19:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h15", "dur_2": "-1h45", "dur_1": "0h15", "dur_0": "-1h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = $720; priority = high.\n\nClaim to verify: \"the status is $720\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p1": "0.150", "p3": "0.050", "p2": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $576. A fee of $178 is charged, then a deposit of $90 arrives. A second fee of $89 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "384", "answer_0": "399", "answer_3": "375", "answer_2": "401"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.125", "p2": "0.250", "p0": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $632. A fee of $124 is charged, then a deposit of $52 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "525", "answer_1": "506", "answer_3": "495", "answer_0": "498"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $515. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, June 04, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 477:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed May 13, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 28", "date_3": "May 24", "date_0": "May 26", "date_1": "May 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = 2026-11-11; priority = high.\n\nClaim to verify: \"the amount is 2026-11-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $15k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $690. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.545", "p2": "0.417", "p0": "0.583"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 934:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, May 03, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1970. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "56", "age_3": "54", "age_1": "55", "age_2": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 209 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "21", "answer_0": "30", "answer_1": "43", "answer_3": "34"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 859:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $389. A fee of $103 is charged, then a deposit of $70 arrives. A second fee of $51 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "305", "answer_3": "324", "answer_1": "310", "answer_2": "307"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Wednesday, June 24, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.600", "p3": "0.250", "p0": "0.400", "p1": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=214, unit_price=796, discount_pct=0, region=EU, sku_total=1010 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "170558", "v2": "17034", "v0": "170344", "v1": "1010"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=657, unit_price=867, discount_pct=0, region=EU, sku_total=1524 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1524", "v2": "56961", "v0": "569619", "v3": "570276"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1919 per year. It increases by 15% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2194.85", "answer_1": "2206.85", "answer_2": "1631.15", "answer_3": "2170.85"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.222", "p1": "0.778", "p2": "0.778", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.429", "p2": "0.500", "p1": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $27k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 842:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 383 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "116", "answer_2": "110", "answer_3": "113", "answer_0": "96"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed May 02, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 05", "date_1": "May 06", "date_0": "May 07", "date_3": "May 08"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = 2026-05-20; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Wednesday, September 23, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 927:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed March 23, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 28", "date_3": "March 30", "date_0": "March 30", "date_1": "March 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = $159; priority = low.\n\nClaim to verify: \"the date is $159\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, June 27, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=755, unit_price=126, discount_pct=0, region=EU, sku_total=881 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "95885", "v0": "95130", "v1": "881", "v2": "9513"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $724. A fee of $90 is charged, then a deposit of $87 arrives. A second fee of $45 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "658", "answer_3": "654", "answer_0": "676", "answer_1": "685"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "15", "answer_1": "23", "answer_0": "33", "answer_2": "18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 269 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "9", "answer_0": "24", "answer_3": "13", "answer_1": "4"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.167", "p0": "0.667", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1960. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "67", "age_2": "66", "age_1": "64", "age_0": "65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.150", "p0": "0.200", "p3": "0.050"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 13:45 to 16:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "1h45", "dur_2": "2h15", "dur_3": "1h30", "dur_1": "2h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 315:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $756. A fee of $127 is charged, then a deposit of $68 arrives. A second fee of $63 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "654", "answer_2": "664", "answer_0": "634", "answer_3": "615"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_0": "40", "answer_1": "32", "answer_3": "10"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=140, unit_price=154, discount_pct=0, region=EU, sku_total=294 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "294", "v3": "21700", "v2": "2156", "v0": "21560"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $339. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = 2026-04-11; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $747. A fee of $153 is charged, then a deposit of $60 arrives. A second fee of $76 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "578", "answer_2": "602", "answer_3": "607", "answer_1": "549"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p2": "0.375", "p0": "0.625", "p3": "0.571"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_1": "24", "answer_3": "18", "answer_0": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 362:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 578:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 369:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "24", "answer_2": "20", "answer_0": "40", "answer_3": "20"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $241. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed October 15, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 23", "date_0": "October 21", "date_1": "October 19", "date_3": "October 21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_0": "35", "answer_2": "21", "answer_1": "23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.556", "p0": "0.444", "p3": "0.375", "p2": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1960. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "67", "age_3": "64", "age_1": "65", "age_0": "66"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1982. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "43", "age_0": "42", "age_1": "41", "age_3": "44"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1963. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "61", "age_2": "62", "age_3": "63", "age_1": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $696. A fee of $49 is charged, then a deposit of $82 arrives. A second fee of $24 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "705", "answer_1": "683", "answer_3": "728", "answer_2": "734"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $259 per year. It increases by 20% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "280.8", "answer_3": "250.8", "answer_1": "310.8", "answer_2": "207.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=486, unit_price=189, discount_pct=0, region=EU, sku_total=675 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "9185", "v0": "91854", "v3": "92340", "v1": "675"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $580. A fee of $61 is charged, then a deposit of $71 arrives. A second fee of $30 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "572", "answer_1": "547", "answer_2": "566", "answer_0": "560"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $834. A fee of $92 is charged, then a deposit of $53 arrives. A second fee of $46 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "769", "answer_0": "749", "answer_3": "744", "answer_1": "736"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1985. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "39", "age_0": "41", "age_1": "40", "age_2": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_0": "32", "answer_3": "16", "answer_1": "18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 443:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=144, unit_price=723, discount_pct=0, region=EU, sku_total=867 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "867", "v2": "10411", "v0": "104112", "v3": "104256"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.375", "p3": "0.125", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p3": "0.125", "p0": "0.250", "p2": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.727", "p1": "0.273", "p3": "0.700", "p2": "0.273"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "91", "answer_1": "49", "answer_2": "42", "answer_3": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = 2026-12-12; priority = high.\n\nClaim to verify: \"the owner is 2026-12-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 08:00 to 11:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h15", "dur_3": "2h00", "dur_1": "3h00", "dur_2": "2h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 5 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $650. A fee of $109 is charged, then a deposit of $64 arrives. A second fee of $54 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "526", "answer_0": "551", "answer_2": "524", "answer_1": "522"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = $618; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 06:45 to 07:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "0h00", "dur_1": "1h00", "dur_0": "0h45", "dur_2": "1h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $605. A fee of $130 is charged, then a deposit of $61 arrives. A second fee of $65 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "482", "answer_0": "471", "answer_1": "481", "answer_2": "455"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = ops; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.667", "p0": "0.333", "p3": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_0": "42", "answer_3": "24", "answer_1": "22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = 2026-10-06; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, September 22, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 392 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "358", "answer_3": "384", "answer_0": "378", "answer_1": "366"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 206 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "202", "answer_1": "203", "answer_0": "216", "answer_2": "228"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.333", "p3": "0.625", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "37", "answer_3": "40", "answer_0": "72", "answer_2": "32"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = 2026-09-04; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 87 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "47", "answer_3": "72", "answer_0": "60", "answer_2": "69"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $13k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=204, unit_price=213, discount_pct=0, region=EU, sku_total=417 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "43452", "v3": "43248", "v1": "417", "v2": "4345"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 8 grams of salt per serving and serves 8 people. A shopper bought a 64 gram box and a 640 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 07:00 to 14:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "8h00", "dur_1": "7h45", "dur_3": "6h45", "dur_0": "7h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed February 14, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 20", "date_1": "February 17", "date_3": "February 19", "date_0": "February 18"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=364, unit_price=275, discount_pct=0, region=EU, sku_total=639 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "100464", "v0": "100100", "v2": "10010", "v1": "639"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1960. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "65", "age_3": "63", "age_2": "66", "age_1": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $496. A fee of $83 is charged, then a deposit of $52 arrives. A second fee of $41 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "409", "answer_0": "424", "answer_2": "403", "answer_3": "439"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = 2026-02-23; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 364 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "200", "answer_3": "206", "answer_0": "210", "answer_1": "212"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, January 07, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p1": "0.150", "p3": "0.050", "p2": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=819, unit_price=290, discount_pct=0, region=EU, sku_total=1109 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "238329", "v2": "23751", "v0": "237510", "v1": "1109"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.462", "p3": "0.500", "p1": "0.462", "p0": "0.538"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.444", "p2": "0.444", "p3": "0.500", "p0": "0.556"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 267 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "138", "answer_1": "141", "answer_3": "155", "answer_0": "144"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 3 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=368, unit_price=891, discount_pct=0, region=EU, sku_total=1259 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1259", "v3": "328256", "v0": "327888", "v2": "32788"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed March 16, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 20", "date_2": "March 22", "date_0": "March 20", "date_3": "March 22"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1109 per year. It increases by 10% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1180.9", "answer_3": "1205.9", "answer_2": "998.1", "answer_1": "1219.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 08:30 to 14:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "5h45", "dur_0": "5h00", "dur_2": "5h15", "dur_3": "4h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, July 28, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "62", "age_0": "64", "age_1": "63", "age_2": "65"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $216. A fee of $109 is charged, then a deposit of $56 arrives. A second fee of $54 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "79", "answer_3": "130", "answer_0": "109", "answer_2": "138"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1965. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "61", "age_2": "62", "age_1": "60", "age_3": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p1": "0.500", "p3": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 998:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = $764; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = $330; priority = low.\n\nClaim to verify: \"the date is $330\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 394 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_2": "60", "answer_3": "91", "answer_1": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p0": "0.500", "p1": "0.625", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.333", "p1": "0.500", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $789. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed January 10, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 18", "date_0": "January 20", "date_1": "January 17", "date_3": "January 19"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = 2026-07-16; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = 2026-08-02; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.636", "p1": "0.636", "p3": "0.300", "p0": "0.364"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1968. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "56", "age_2": "57", "age_3": "54", "age_1": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $400. A fee of $193 is charged, then a deposit of $66 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "177", "answer_3": "194", "answer_2": "173", "answer_1": "205"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $255 per year. It increases by 10% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "234.5", "answer_2": "229.5", "answer_1": "280.5", "answer_0": "246.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p1": "0.333", "p3": "0.600", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1983. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "40", "age_3": "43", "age_2": "42", "age_0": "41"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $393. A fee of $74 is charged, then a deposit of $89 arrives. A second fee of $37 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "354", "answer_3": "366", "answer_0": "371", "answer_2": "380"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1985. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "40", "age_3": "38", "age_1": "39", "age_2": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = $67; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, July 11, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed June 28, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 08", "date_2": "July 10", "date_1": "July 06", "date_3": "July 08"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1187 per year. It increases by 5% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1127.65", "answer_0": "1213.35", "answer_3": "1175.35", "answer_1": "1246.35"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = 2026-04-22; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 205:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1681 per year. It increases by 25% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2090.25", "answer_0": "2055.25", "answer_1": "2101.25", "answer_2": "1260.75"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.100", "p2": "0.050", "p1": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 616:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = $875; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.222", "p2": "0.222", "p0": "0.778", "p3": "0.750"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 296 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "70", "answer_3": "86", "answer_1": "83", "answer_2": "62"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $8k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed December 19, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 23", "date_0": "December 24", "date_3": "December 25", "date_2": "December 26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = ops; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $825. A fee of $190 is charged, then a deposit of $66 arrives. A second fee of $95 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "628", "answer_0": "606", "answer_2": "614", "answer_3": "627"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = pending; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=595, unit_price=428, discount_pct=0, region=EU, sku_total=1023 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "254660", "v2": "25466", "v3": "255255", "v1": "1023"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_1": "29", "answer_0": "54", "answer_3": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1981. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "42", "age_0": "43", "age_2": "44", "age_3": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 171 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "32", "answer_2": "22", "answer_1": "44", "answer_3": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 84 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "108", "answer_1": "106", "answer_3": "115", "answer_2": "96"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = ops; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 105 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "28", "answer_0": "24", "answer_2": "37", "answer_1": "36"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, October 18, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1091 per year. It increases by 10% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1200.1", "answer_3": "1177.1", "answer_0": "1159.1", "answer_2": "981.9"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.250", "p2": "0.150", "p0": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 834:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 340:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "10", "answer_3": "10", "answer_0": "20", "answer_1": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 866:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = 2026-07-21; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.462", "p3": "0.500", "p1": "0.462", "p0": "0.538"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $25k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.250", "p1": "0.417", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 244:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, March 09, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p3": "0.333", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1963. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "63", "age_1": "60", "age_2": "62", "age_0": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed October 12, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 23", "date_1": "October 21", "date_2": "October 21", "date_0": "October 23"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.467", "p2": "0.467", "p3": "0.500", "p0": "0.533"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 314:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.333", "p1": "0.500", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.286", "p3": "0.167", "p1": "0.714", "p2": "0.714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $309. A fee of $181 is charged, then a deposit of $49 arrives. A second fee of $90 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "87", "answer_2": "116", "answer_3": "110", "answer_1": "114"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 887:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.500", "p3": "0.167", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, November 11, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_3": "40", "answer_1": "29", "answer_0": "64"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 205 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "70", "answer_1": "103", "answer_0": "84", "answer_3": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p0": "0.667", "p2": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 648:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.083", "p0": "0.333", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $774. A fee of $63 is charged, then a deposit of $30 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "710", "answer_2": "685", "answer_3": "716", "answer_1": "740"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 389 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "63", "answer_1": "81", "answer_3": "66", "answer_2": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 89 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "270", "answer_3": "267", "answer_1": "272", "answer_0": "252"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_0": "44", "answer_2": "20", "answer_1": "26"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = 2026-02-26; priority = high.\n\nClaim to verify: \"the date is 2026-02-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1306 per year. It increases by 25% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1632.5", "answer_0": "1599.5", "answer_3": "1593.5", "answer_2": "979.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $819. A fee of $59 is charged, then a deposit of $58 arrives. A second fee of $29 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "801", "answer_3": "765", "answer_1": "767", "answer_0": "789"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $495. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Sunday, September 06, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $134. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = 2026-06-19; priority = high.\n\nClaim to verify: \"the owner is 2026-06-19\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 4 grams of salt per serving and serves 6 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, September 09, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 9 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 367 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_3": "27", "answer_2": "32", "answer_1": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=976, unit_price=939, discount_pct=0, region=EU, sku_total=1915 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "915488", "v0": "916464", "v2": "91646", "v1": "1915"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $209 per year. It increases by 15% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "204.35", "answer_1": "240.35", "answer_2": "177.65", "answer_3": "216.35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, November 18, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 12 people. A shopper bought a 108 gram box and a 1080 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.500", "p1": "0.833", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1173 per year. It increases by 5% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1218.65", "answer_2": "1114.35", "answer_1": "1231.65", "answer_0": "1201.65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_1": "33", "answer_3": "18", "answer_0": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1979. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "47", "age_0": "46", "age_1": "45", "age_3": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1166 per year. It increases by 5% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1224.3", "answer_0": "1189.3", "answer_3": "1183.3", "answer_2": "1107.7"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 4 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A shift runs from 11:45 to 14:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "2h45", "dur_3": "1h30", "dur_1": "2h30", "dur_0": "2h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = closed; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.083", "p0": "0.167", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 788:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, March 16, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $23k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.636", "p3": "0.600", "p1": "0.364", "p2": "0.364"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1963. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "62", "age_0": "63", "age_3": "65", "age_2": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p3": "0.167", "p0": "0.500", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 133 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "94", "answer_2": "101", "answer_0": "112", "answer_1": "118"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1693 per year. It increases by 5% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1759.65", "answer_1": "1777.65", "answer_2": "1608.35", "answer_3": "1737.65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 4 people. A shopper bought a 8 gram box and a 80 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 278 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "126", "answer_2": "106", "answer_3": "118", "answer_1": "112"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 198 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_2": "66", "answer_1": "84", "answer_3": "86"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1985. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "40", "age_3": "37", "age_1": "38", "age_0": "39"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Saturday, September 26, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1966. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "58", "age_0": "59", "age_2": "60", "age_3": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed December 23, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 30", "date_2": "January 01", "date_1": "December 28", "date_3": "December 30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $554. A fee of $176 is charged, then a deposit of $41 arrives. A second fee of $88 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "314", "answer_1": "354", "answer_3": "333", "answer_0": "331"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.250", "p3": "0.083", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 184 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "39", "answer_0": "42", "answer_2": "57", "answer_1": "54"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=775, unit_price=462, discount_pct=0, region=EU, sku_total=1237 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "358050", "v1": "1237", "v3": "358825", "v2": "35805"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $19k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = ops; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 249 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_2": "130", "answer_1": "123", "answer_3": "118"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed June 08, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "June 18", "date_1": "June 16", "date_3": "June 18", "date_2": "June 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = $13; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 826:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = $557; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, October 13, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = 2026-12-12; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p3": "0.333", "p2": "0.600", "p1": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.375", "p0": "0.250", "p2": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 204 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "115", "answer_1": "122", "answer_0": "112", "answer_2": "93"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 187 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "32", "answer_2": "42", "answer_1": "51", "answer_3": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $732. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $33k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $492. A fee of $95 is charged, then a deposit of $64 arrives. A second fee of $47 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "386", "answer_3": "432", "answer_0": "414", "answer_2": "416"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1978. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "47", "age_2": "48", "age_1": "46", "age_3": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = 2026-05-13; priority = high.\n\nClaim to verify: \"the date is 2026-05-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1912 per year. It increases by 25% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2353", "answer_1": "2390", "answer_3": "2342", "answer_2": "1434"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 324:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = $507; priority = high.\n\nClaim to verify: \"the date is $507\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 5 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 764:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.333", "p3": "0.250", "p2": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 292 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_3": "49", "answer_2": "50", "answer_1": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 50 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "93", "answer_1": "75", "answer_0": "90", "answer_2": "74"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "88", "answer_1": "53", "answer_2": "48", "answer_3": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $690. A fee of $129 is charged, then a deposit of $88 arrives. A second fee of $64 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "577", "answer_3": "568", "answer_2": "577", "answer_0": "585"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, April 21, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.750", "p1": "0.750", "p0": "0.250", "p3": "0.143"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $14k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 11:45 to 19:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "6h30", "dur_2": "7h00", "dur_1": "7h30", "dur_0": "6h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p3": "0.200", "p2": "0.667", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = 2026-11-01; priority = low.\n\nClaim to verify: \"the date is 2026-11-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 113 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "24", "answer_2": "56", "answer_3": "31", "answer_0": "40"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, December 21, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $9k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $748. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 174 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "225", "answer_3": "237", "answer_2": "235", "answer_1": "209"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.364", "p2": "0.583", "p1": "0.583", "p0": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1971. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "53", "age_1": "52", "age_3": "51", "age_2": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.250", "p2": "0.083", "p0": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $487 per year. It increases by 20% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "584.4", "answer_3": "579.4", "answer_2": "389.6", "answer_0": "557.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = $256; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, October 15, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed August 09, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "August 20", "date_3": "August 18", "date_1": "August 16", "date_0": "August 18"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $604. A fee of $37 is charged, then a deposit of $33 arrives. A second fee of $18 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "609", "answer_0": "582", "answer_3": "594", "answer_1": "609"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1971. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "55", "age_1": "54", "age_2": "56", "age_3": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed March 11, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 17", "date_2": "March 19", "date_0": "March 17", "date_1": "March 15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 6 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 06:45 to 12:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "5h00", "dur_1": "5h15", "dur_2": "5h15", "dur_3": "4h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Thursday, December 24, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 288 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "104", "answer_1": "103", "answer_0": "108", "answer_3": "116"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed September 04, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 17", "date_2": "September 15", "date_1": "September 13", "date_3": "September 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.375", "p3": "0.125", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=984, unit_price=961, discount_pct=0, region=EU, sku_total=1945 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1945", "v0": "945624", "v2": "94562", "v3": "946608"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1989. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "37", "age_3": "35", "age_1": "36", "age_2": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 27, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 03", "date_3": "May 05", "date_0": "May 05", "date_1": "May 03"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed June 11, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 18", "date_1": "June 16", "date_2": "June 16", "date_0": "June 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $645; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.385", "p2": "0.571", "p1": "0.571", "p0": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = pending; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $36k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1622 per year. It increases by 20% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1946.4", "answer_3": "1943.4", "answer_0": "1925.4", "answer_2": "1297.6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $599 per year. It increases by 25% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "719.75", "answer_2": "449.25", "answer_1": "748.75", "answer_0": "706.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=815, unit_price=232, discount_pct=0, region=EU, sku_total=1047 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1047", "v3": "188265", "v2": "18908", "v0": "189080"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 333 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "132", "answer_3": "121", "answer_1": "147", "answer_0": "140"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 320 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "8", "answer_1": "20", "answer_2": "42", "answer_0": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $399 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "349.95", "answer_2": "379.05", "answer_0": "386.95", "answer_1": "418.95"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.500", "p1": "0.833", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 10:30 to 14:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "2h30", "dur_0": "2h45", "dur_2": "3h00", "dur_1": "3h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1981. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "45", "age_2": "44", "age_0": "43", "age_1": "42"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1962. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "64", "age_1": "63", "age_2": "65", "age_3": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 156 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "100", "answer_2": "118", "answer_1": "88", "answer_3": "82"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.167", "p0": "0.250", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 15:15 to 18:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h15", "dur_0": "3h00", "dur_3": "2h30", "dur_1": "3h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = $863; priority = high.\n\nClaim to verify: \"the date is $863\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.167", "p1": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed August 11, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 16", "date_1": "August 14", "date_0": "August 14", "date_2": "August 12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $433. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 406:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed May 24, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 01", "date_2": "June 03", "date_1": "May 30", "date_0": "June 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed May 17, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 25", "date_0": "May 27", "date_1": "May 25", "date_3": "May 27"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.167", "p3": "0.083", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p2": "0.571", "p0": "0.429", "p3": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 6 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 17:30 to 21:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "2h30", "dur_2": "3h15", "dur_1": "3h30", "dur_0": "2h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=505, unit_price=340, discount_pct=0, region=EU, sku_total=845 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "171700", "v1": "845", "v2": "17170", "v3": "172205"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $297. A fee of $75 is charged, then a deposit of $45 arrives. A second fee of $37 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "230", "answer_2": "236", "answer_3": "226", "answer_1": "232"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p0": "0.333", "p1": "0.667", "p3": "0.273"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=520, unit_price=286, discount_pct=0, region=EU, sku_total=806 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "14872", "v1": "806", "v0": "148720", "v3": "148200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=811, unit_price=120, discount_pct=0, region=EU, sku_total=931 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "931", "v3": "96509", "v2": "9732", "v0": "97320"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = 2026-04-09; priority = high.\n\nClaim to verify: \"the owner is 2026-04-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 06:15 to 08:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "1h00", "dur_1": "2h00", "dur_0": "1h45", "dur_2": "2h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = 2026-06-24; priority = high.\n\nClaim to verify: \"the amount is 2026-06-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 3 grams of salt per serving and serves 9 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p1": "0.250", "p3": "0.083", "p2": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p1": "0.250", "p3": "0.083", "p2": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1967 per year. It increases by 25% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2417.75", "answer_1": "2458.75", "answer_2": "1475.25", "answer_3": "2438.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 18:45 to 22:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h45", "dur_3": "2h30", "dur_1": "3h30", "dur_2": "3h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed August 27, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 10", "date_1": "September 04", "date_3": "September 06", "date_0": "September 08"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p2": "0.200", "p0": "0.800", "p3": "0.778"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, April 27, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.700", "p0": "0.300", "p1": "0.700", "p3": "0.222"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = 2026-08-28; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed June 10, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 26", "date_0": "June 24", "date_1": "June 20", "date_3": "June 22"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1970. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "57", "age_3": "58", "age_0": "56", "age_1": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 290 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "137", "answer_3": "123", "answer_1": "113", "answer_0": "128"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.462", "p2": "0.500", "p1": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.615", "p3": "0.333", "p0": "0.385", "p1": "0.615"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = $678; priority = high.\n\nClaim to verify: \"the owner is $678\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed September 23, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "October 01", "date_2": "October 03", "date_1": "September 29", "date_3": "October 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "28", "answer_3": "35", "answer_1": "33", "answer_0": "63"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, March 01, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $796. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $766. A fee of $69 is charged, then a deposit of $54 arrives. A second fee of $34 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "723", "answer_2": "743", "answer_0": "717", "answer_3": "724"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 837:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 5 grams of salt per serving and serves 3 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.500", "p1": "0.667", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 09:30 to 19:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "9h15", "dur_1": "9h30", "dur_2": "9h45", "dur_3": "8h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = 2026-12-10; priority = low.\n\nClaim to verify: \"the amount is 2026-12-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = $611; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed May 17, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 19", "date_3": "May 23", "date_0": "May 21", "date_1": "May 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 288 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "164", "answer_0": "144", "answer_3": "156", "answer_1": "150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 51 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "89", "answer_0": "80", "answer_3": "70", "answer_1": "93"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = $173; priority = high.\n\nClaim to verify: \"the status is $173\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 6 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A shift runs from 14:45 to 22:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "6h15", "dur_2": "7h00", "dur_1": "7h15", "dur_0": "6h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed April 18, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 28", "date_0": "April 29", "date_2": "April 27", "date_1": "April 26"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 163 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "134", "answer_0": "120", "answer_3": "113", "answer_2": "126"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.125", "p3": "0.125", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 129 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_3": "125", "answer_2": "128", "answer_1": "125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = 2026-09-24; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: status = $732; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $23k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $839. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed March 12, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 16", "date_3": "March 18", "date_2": "March 20", "date_0": "March 18"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, May 15, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $530. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, April 27, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = 2026-02-25; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed July 03, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 16", "date_1": "July 10", "date_0": "July 14", "date_3": "July 12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 187 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "141", "answer_0": "125", "answer_2": "115", "answer_3": "130"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p1": "0.333", "p2": "0.333", "p3": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $22k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $864. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.667", "p3": "0.200", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.429", "p3": "0.333", "p2": "0.571", "p1": "0.571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=369, unit_price=890, discount_pct=0, region=EU, sku_total=1259 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "328410", "v2": "32841", "v3": "328779", "v1": "1259"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.462", "p2": "0.538", "p3": "0.417", "p1": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1549 per year. It increases by 20% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1239.2", "answer_1": "1858.8", "answer_3": "1790.8", "answer_0": "1829.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1981. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "43", "age_2": "46", "age_1": "44", "age_0": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 9 grams of salt per serving and serves 11 people. A shopper bought a 99 gram box and a 990 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, September 01, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.222", "p2": "0.778", "p1": "0.778", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.556", "p3": "0.375", "p1": "0.556", "p0": "0.444"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=359, unit_price=236, discount_pct=0, region=EU, sku_total=595 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "8472", "v3": "84365", "v1": "595", "v0": "84724"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_3": "12", "answer_1": "14", "answer_0": "24"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1958. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "66", "age_0": "67", "age_2": "68", "age_3": "69"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 201 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "84", "answer_1": "90", "answer_3": "99", "answer_2": "104"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "14", "answer_0": "42", "answer_3": "28", "answer_1": "18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = $859; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1983. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "44", "age_1": "42", "age_0": "43", "age_3": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "15", "answer_1": "17", "answer_3": "6", "answer_0": "21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 172 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "299", "answer_0": "315", "answer_3": "330", "answer_2": "311"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.750", "p1": "0.750", "p3": "0.143", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = open; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.400", "p1": "0.400", "p3": "0.500", "p0": "0.600"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 541:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $491. A fee of $167 is charged, then a deposit of $77 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "318", "answer_3": "333", "answer_2": "347", "answer_1": "341"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = 2026-12-22; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1967. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "57", "age_0": "58", "age_2": "59", "age_3": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.636", "p0": "0.667", "p1": "0.333", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 212 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "15", "answer_0": "24", "answer_3": "33", "answer_1": "40"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_0": "50", "answer_1": "34", "answer_3": "20"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $650. A fee of $41 is charged, then a deposit of $22 arrives. A second fee of $20 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "622", "answer_3": "587", "answer_0": "611", "answer_1": "627"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, November 14, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 77 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "272", "answer_0": "270", "answer_3": "262", "answer_1": "258"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.636", "p2": "0.364", "p3": "0.600", "p1": "0.364"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1978. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "44", "age_1": "45", "age_0": "46", "age_2": "47"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1982. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "42", "age_2": "44", "age_0": "43", "age_3": "45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 08:15 to 14:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "6h30", "dur_0": "6h00", "dur_1": "6h30", "dur_3": "5h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed February 14, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 22", "date_0": "February 23", "date_2": "February 21", "date_1": "February 20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p3": "0.400", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = 2026-01-13; priority = high.\n\nClaim to verify: \"the owner is 2026-01-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.444", "p1": "0.444", "p0": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1967. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "59", "age_1": "57", "age_0": "58", "age_3": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 292:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, November 17, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.250", "p0": "0.375", "p3": "0.125"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1980. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "46", "age_3": "48", "age_2": "47", "age_1": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=824, unit_price=780, discount_pct=0, region=EU, sku_total=1604 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "64272", "v3": "643544", "v0": "642720", "v1": "1604"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "32", "answer_1": "34", "answer_0": "48", "answer_3": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p3": "0.050", "p1": "0.200", "p0": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 5 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 12:30 to 13:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "1h00", "dur_2": "1h00", "dur_0": "0h45", "dur_3": "0h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 3 grams of salt per serving and serves 11 people. A shopper bought a 33 gram box and a 330 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "20", "answer_1": "21", "answer_0": "36", "answer_2": "16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=965, unit_price=329, discount_pct=0, region=EU, sku_total=1294 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "318450", "v0": "317485", "v2": "31748", "v1": "1294"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = closed; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 07:00 to 22:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "14h30", "dur_1": "15h15", "dur_2": "14h45", "dur_3": "14h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 429:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $812. A fee of $55 is charged, then a deposit of $74 arrives. A second fee of $27 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "798", "answer_0": "804", "answer_3": "823", "answer_1": "820"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.250", "p1": "0.417", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed April 01, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 09", "date_3": "April 11", "date_0": "April 13", "date_2": "April 15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 9 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 764:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $788. A fee of $59 is charged, then a deposit of $32 arrives. A second fee of $29 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "730", "answer_0": "732", "answer_2": "755", "answer_1": "746"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1961. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "64", "age_3": "67", "age_2": "66", "age_0": "65"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = 2026-06-25; priority = high.\n\nClaim to verify: \"the date is 2026-06-25\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Tuesday, March 10, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 08:30 to 11:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h15", "dur_3": "2h00", "dur_1": "3h00", "dur_2": "2h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = 2026-02-16; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $800. A fee of $65 is charged, then a deposit of $72 arrives. A second fee of $32 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "775", "answer_1": "753", "answer_3": "756", "answer_2": "793"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=595, unit_price=404, discount_pct=0, region=EU, sku_total=999 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "240975", "v1": "999", "v2": "24038", "v0": "240380"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 3 grams of salt per serving and serves 3 people. A shopper bought a 9 gram box and a 90 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.500", "p3": "0.333", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = 2026-01-05; priority = high.\n\nClaim to verify: \"the amount is 2026-01-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 06:00 to 17:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "11h15", "dur_2": "11h00", "dur_3": "10h15", "dur_0": "10h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1979. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "46", "age_0": "47", "age_3": "45", "age_2": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Wednesday, October 07, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 4 people. A shopper bought a 8 gram box and a 80 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = 2026-05-13; priority = high.\n\nClaim to verify: \"the amount is 2026-05-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.625", "p0": "0.375", "p3": "0.286", "p1": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p2": "0.417", "p0": "0.583", "p3": "0.545"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_3": "12", "answer_0": "30", "answer_1": "22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $273. A fee of $62 is charged, then a deposit of $24 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "204", "answer_2": "232", "answer_1": "228", "answer_3": "219"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 7 grams of salt per serving and serves 4 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $688 per year. It increases by 25% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "841", "answer_0": "824", "answer_1": "860", "answer_2": "516"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 284 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "378", "answer_2": "366", "answer_3": "376", "answer_1": "372"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 8 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 361 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "94", "answer_0": "96", "answer_3": "108", "answer_2": "101"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, May 21, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1229 per year. It increases by 20% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "983.2", "answer_1": "1474.8", "answer_3": "1411.8", "answer_0": "1439.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $867 per year. It increases by 10% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "953.7", "answer_3": "915.7", "answer_2": "780.3", "answer_0": "929.7"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1979. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "48", "age_1": "46", "age_0": "47", "age_3": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.429", "p0": "0.571", "p3": "0.500", "p1": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.167", "p0": "0.333", "p1": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed March 15, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 22", "date_2": "March 22", "date_0": "March 20", "date_1": "March 20"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $916 per year. It increases by 15% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1053.4", "answer_0": "1006.4", "answer_3": "980.4", "answer_2": "778.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $495. A fee of $44 is charged, then a deposit of $40 arrives. A second fee of $22 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "466", "answer_0": "469", "answer_1": "466", "answer_2": "487"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.727", "p1": "0.273", "p2": "0.273", "p3": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p3": "0.125", "p0": "0.250", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1971. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "56", "age_0": "54", "age_1": "53", "age_2": "55"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = ops; priority = low.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 9 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "30", "answer_3": "20", "answer_1": "14", "answer_2": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=257, unit_price=489, discount_pct=0, region=EU, sku_total=746 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "12567", "v0": "125673", "v3": "125930", "v1": "746"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A recipe needs 2 grams of salt per serving and serves 9 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1980. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "45", "age_3": "44", "age_0": "46", "age_2": "47"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 20:45 to 21:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "0h15", "dur_2": "0h30", "dur_3": "0h00", "dur_1": "1h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.538", "p2": "0.462", "p1": "0.462", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = $732; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=367, unit_price=178, discount_pct=0, region=EU, sku_total=545 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "545", "v3": "65693", "v0": "65326", "v2": "6532"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_3": "16", "answer_0": "40", "answer_1": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Wednesday, October 21, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 985:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, February 19, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1962. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "66", "age_1": "63", "age_0": "64", "age_2": "65"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 757:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $254 per year. It increases by 20% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "203.2", "answer_3": "296.8", "answer_0": "275.8", "answer_1": "304.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.700", "p0": "0.300", "p3": "0.222", "p1": "0.700"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1801 per year. It increases by 20% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1440.8", "answer_0": "2111.2", "answer_1": "2161.2", "answer_3": "2124.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=957, unit_price=315, discount_pct=0, region=EU, sku_total=1272 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "300498", "v2": "30145", "v0": "301455", "v1": "1272"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 171 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "79", "answer_2": "103", "answer_0": "90", "answer_3": "84"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_3": "28", "answer_1": "15", "answer_2": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = $804; priority = high.\n\nClaim to verify: \"the owner is $804\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $765. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $789 per year. It increases by 5% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "828.45", "answer_2": "749.55", "answer_3": "757.45", "answer_0": "797.45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.273", "p1": "0.727", "p2": "0.727", "p3": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 374 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "28", "answer_0": "48", "answer_1": "44", "answer_3": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed October 14, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 25", "date_1": "October 23", "date_0": "October 27", "date_3": "October 25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 271 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "54", "answer_2": "58", "answer_1": "37", "answer_3": "62"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p1": "0.667", "p3": "0.250", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = pending; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 269 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "116", "answer_2": "110", "answer_0": "108", "answer_1": "111"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed September 16, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 22", "date_0": "September 22", "date_1": "September 20", "date_2": "September 20"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.385", "p3": "0.583", "p0": "0.615", "p1": "0.385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1924 per year. It increases by 25% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2359", "answer_3": "2350", "answer_1": "2405", "answer_2": "1443"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1975. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "50", "age_2": "52", "age_0": "51", "age_3": "49"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=321, unit_price=139, discount_pct=0, region=EU, sku_total=460 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "4461", "v3": "44940", "v0": "44619", "v1": "460"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 744:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $8k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1618 per year. It increases by 20% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1294.4", "answer_1": "1941.6", "answer_0": "1923.6", "answer_3": "1885.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $487; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 5 people. A shopper bought a 10 gram box and a 100 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.750", "p1": "0.250", "p3": "0.714", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = $743; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.200", "p1": "0.250", "p2": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "36", "answer_2": "30", "answer_0": "66", "answer_3": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 6 grams of salt per serving and serves 12 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 51 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "120", "answer_1": "87", "answer_0": "100", "answer_2": "98"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.167", "p0": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 635:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = 2026-04-04; priority = high.\n\nClaim to verify: \"the status is 2026-04-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.333", "p3": "0.083", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed January 17, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 20", "date_3": "January 23", "date_0": "January 22", "date_1": "January 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $260. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1982. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "44", "age_2": "43", "age_1": "41", "age_0": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $457. A fee of $141 is charged, then a deposit of $75 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "324", "answer_3": "328", "answer_1": "349", "answer_0": "321"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 628:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.455", "p2": "0.500", "p1": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 5 grams of salt per serving and serves 10 people. A shopper bought a 50 gram box and a 500 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p1": "0.333", "p3": "0.636", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, April 14, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1495 per year. It increases by 25% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1868.75", "answer_3": "1819.75", "answer_0": "1834.75", "answer_2": "1121.25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed December 19, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 27", "date_2": "December 30", "date_0": "December 28", "date_1": "December 25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = ops; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, August 21, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_3": "42", "answer_1": "19", "answer_0": "54"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1964. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "62", "age_1": "61", "age_3": "64", "age_2": "63"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 234 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "132", "answer_2": "108", "answer_0": "112", "answer_1": "120"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, June 24, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.429", "p2": "0.533", "p0": "0.467", "p1": "0.533"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=881, unit_price=195, discount_pct=0, region=EU, sku_total=1076 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "171795", "v3": "170914", "v1": "1076", "v2": "17179"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 565:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=653, unit_price=367, discount_pct=0, region=EU, sku_total=1020 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1020", "v2": "23965", "v3": "238998", "v0": "239651"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_1": "27", "answer_0": "48", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "35", "answer_0": "49", "answer_3": "14", "answer_1": "37"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = 2026-03-08; priority = high.\n\nClaim to verify: \"the date is 2026-03-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: date = $359; priority = high.\n\nClaim to verify: \"the date is $359\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1962. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "62", "age_2": "63", "age_1": "61", "age_3": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = $525; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 18:30 to 20:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "1h30", "dur_3": "1h00", "dur_2": "1h45", "dur_1": "2h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.444", "p3": "0.500", "p1": "0.444", "p0": "0.556"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 262:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $538. A fee of $38 is charged, then a deposit of $66 arrives. A second fee of $19 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "547", "answer_1": "574", "answer_2": "545", "answer_3": "544"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_2": "9", "answer_0": "27", "answer_1": "15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $397. A fee of $186 is charged, then a deposit of $31 arrives. A second fee of $93 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "138", "answer_2": "171", "answer_3": "119", "answer_0": "149"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p2": "0.250", "p1": "0.500", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=279, unit_price=305, discount_pct=0, region=EU, sku_total=584 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "84816", "v0": "85095", "v2": "8509", "v1": "584"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, December 09, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, July 02, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1975. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "50", "age_1": "48", "age_3": "51", "age_0": "49"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 768:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $824. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $505. A fee of $85 is charged, then a deposit of $87 arrives. A second fee of $42 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "485", "answer_3": "449", "answer_0": "465", "answer_1": "486"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, February 20, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.429", "p2": "0.500", "p0": "0.500", "p1": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 07:15 to 18:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "11h00", "dur_2": "10h30", "dur_0": "10h15", "dur_3": "10h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = 2026-04-01; priority = high.\n\nClaim to verify: \"the amount is 2026-04-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.778", "p2": "0.222", "p3": "0.750", "p1": "0.222"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1962. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "65", "age_3": "62", "age_1": "63", "age_0": "64"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=603, unit_price=666, discount_pct=0, region=EU, sku_total=1269 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "402201", "v2": "40159", "v1": "1269", "v0": "401598"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, March 12, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = closed; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p0": "0.250", "p1": "0.375", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $791. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 7 grams of salt per serving and serves 3 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, November 12, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=594, unit_price=601, discount_pct=0, region=EU, sku_total=1195 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "356994", "v3": "357588", "v2": "35699", "v1": "1195"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $3k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $370. A fee of $117 is charged, then a deposit of $52 arrives. A second fee of $58 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "222", "answer_3": "230", "answer_1": "221", "answer_0": "247"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed September 27, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 07", "date_2": "October 11", "date_3": "October 09", "date_0": "October 09"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 854:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1980. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "43", "age_3": "46", "age_0": "44", "age_2": "45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=223, unit_price=321, discount_pct=0, region=EU, sku_total=544 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "7158", "v1": "544", "v0": "71583", "v3": "71360"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = $987; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p0": "0.250", "p3": "0.125", "p1": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1962. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "64", "age_2": "65", "age_1": "63", "age_3": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 10:30 to 15:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "5h00", "dur_2": "5h00", "dur_0": "4h30", "dur_3": "4h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "15", "answer_1": "13", "answer_2": "10", "answer_0": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 374:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, December 28, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: status = $465; priority = low.\n\nClaim to verify: \"the status is $465\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = $782; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $367 per year. It increases by 20% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "440.4", "answer_0": "397.4", "answer_3": "371.4", "answer_2": "293.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, August 10, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = $80; priority = low.\n\nClaim to verify: \"the owner is $80\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Monday, January 26, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 6 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_3": "24", "answer_1": "51", "answer_2": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed March 13, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 23", "date_0": "March 27", "date_3": "March 25", "date_2": "March 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.429", "p1": "0.571", "p2": "0.571", "p3": "0.385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=749, unit_price=528, discount_pct=0, region=EU, sku_total=1277 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "395472", "v3": "394723", "v1": "1277", "v2": "39547"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = 2026-02-11; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $14k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 990:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1619 per year. It increases by 25% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1214.25", "answer_1": "2023.75", "answer_3": "1964.75", "answer_0": "1981.75"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_0": "42", "answer_1": "32", "answer_2": "30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = $376; priority = high.\n\nClaim to verify: \"the status is $376\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1976. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "51", "age_0": "50", "age_1": "49", "age_3": "52"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $848 per year. It increases by 10% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "886.8", "answer_1": "932.8", "answer_2": "763.2", "answer_3": "903.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 8 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed November 27, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 02", "date_1": "December 02", "date_0": "December 04", "date_3": "December 04"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1990. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "33", "age_3": "36", "age_0": "34", "age_2": "35"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.222", "p0": "0.778", "p2": "0.222", "p3": "0.750"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 124 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "97", "answer_2": "91", "answer_0": "84", "answer_3": "69"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = $703; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=676, unit_price=643, discount_pct=0, region=EU, sku_total=1319 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1319", "v3": "433992", "v0": "434668", "v2": "43466"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = 2026-10-27; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 600:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 313 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "80", "answer_0": "60", "answer_1": "71", "answer_3": "74"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, December 03, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1969. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "53", "age_1": "54", "age_0": "55", "age_2": "56"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $586. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.583", "p3": "0.364", "p1": "0.583", "p0": "0.417"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $569. A fee of $100 is charged, then a deposit of $22 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "441", "answer_2": "438", "answer_3": "466", "answer_1": "431"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed August 15, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 25", "date_3": "August 27", "date_2": "August 30", "date_0": "August 28"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed March 11, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 17", "date_2": "March 19", "date_0": "March 17", "date_1": "March 15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1359 per year. It increases by 15% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1516.85", "answer_3": "1552.85", "answer_1": "1562.85", "answer_2": "1155.15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1990. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "37", "age_1": "34", "age_2": "36", "age_0": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $12k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed August 26, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "August 29", "date_0": "August 31", "date_1": "August 29", "date_3": "August 31"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, December 20, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.571", "p0": "0.429", "p3": "0.333", "p1": "0.571"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = 2026-09-04; priority = high.\n\nClaim to verify: \"the owner is 2026-09-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.250", "p1": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $605. A fee of $168 is charged, then a deposit of $59 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "439", "answer_3": "432", "answer_2": "415", "answer_0": "412"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = open; priority = high.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p0": "0.667", "p3": "0.167", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = 2026-12-07; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1110 per year. It increases by 5% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1054.5", "answer_1": "1165.5", "answer_0": "1117.5", "answer_3": "1137.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 801:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = $193; priority = high.\n\nClaim to verify: \"the owner is $193\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 216 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "126", "answer_3": "112", "answer_1": "120", "answer_2": "120"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $499 per year. It increases by 5% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "488.95", "answer_3": "501.95", "answer_1": "523.95", "answer_2": "474.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Friday, August 21, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.125", "p1": "0.375", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = ops; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $210. A fee of $152 is charged, then a deposit of $31 arrives. A second fee of $76 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "7", "answer_3": "0", "answer_2": "-7", "answer_0": "13"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p3": "0.600", "p0": "0.667", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=354, unit_price=523, discount_pct=0, region=EU, sku_total=877 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "18514", "v0": "185142", "v1": "877", "v3": "184788"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, August 25, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.636", "p1": "0.636", "p3": "0.300", "p0": "0.364"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=980, unit_price=891, discount_pct=0, region=EU, sku_total=1871 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1871", "v3": "874160", "v2": "87318", "v0": "873180"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p3": "0.050", "p2": "0.050", "p1": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1986. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "40", "age_3": "41", "age_0": "39", "age_1": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed February 07, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 15", "date_0": "February 17", "date_1": "February 14", "date_3": "February 16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "84", "answer_1": "48", "answer_3": "42", "answer_2": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.417", "p3": "0.083", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed January 02, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 13", "date_0": "January 15", "date_1": "January 11", "date_3": "January 13"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $405 per year. It increases by 10% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "397.5", "answer_2": "364.5", "answer_3": "378.5", "answer_1": "445.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $608. A fee of $140 is charged, then a deposit of $52 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "476", "answer_2": "439", "answer_0": "450", "answer_1": "433"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $32k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1975. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "48", "age_0": "49", "age_3": "47", "age_2": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 151 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "110", "answer_3": "131", "answer_0": "112", "answer_2": "122"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1984. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "40", "age_2": "42", "age_0": "41", "age_3": "39"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $334. A fee of $116 is charged, then a deposit of $66 arrives. A second fee of $58 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "238", "answer_3": "248", "answer_1": "202", "answer_0": "226"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 457:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed December 16, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 25", "date_0": "December 25", "date_2": "December 23", "date_1": "December 23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=527, unit_price=965, discount_pct=0, region=EU, sku_total=1492 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1492", "v3": "508028", "v0": "508555", "v2": "50855"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, December 09, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Thursday, May 28, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = 2026-10-16; priority = high.\n\nClaim to verify: \"the owner is 2026-10-16\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $354. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Saturday, May 09, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, June 09, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 9 grams of salt per serving and serves 3 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = $500; priority = high.\n\nClaim to verify: \"the amount is $500\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $200. A fee of $141 is charged, then a deposit of $41 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "5", "answer_0": "30", "answer_1": "56", "answer_2": "41"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 7 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $281. A fee of $134 is charged, then a deposit of $77 arrives. A second fee of $67 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "155", "answer_3": "144", "answer_1": "181", "answer_0": "157"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=626, unit_price=475, discount_pct=0, region=EU, sku_total=1101 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "297350", "v3": "296724", "v1": "1101", "v2": "29735"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1979. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "44", "age_2": "46", "age_3": "47", "age_0": "45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=213, unit_price=767, discount_pct=0, region=EU, sku_total=980 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "163371", "v1": "980", "v2": "16337", "v3": "163584"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 524:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.375", "p1": "0.556", "p0": "0.444", "p2": "0.556"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1362 per year. It increases by 15% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1571.3", "answer_0": "1545.3", "answer_1": "1566.3", "answer_2": "1157.7"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, January 24, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1788 per year. It increases by 25% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2235", "answer_2": "1341", "answer_0": "2188", "answer_3": "2152"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.375", "p3": "0.125", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 461:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $218. A fee of $93 is charged, then a deposit of $57 arrives. A second fee of $46 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "136", "answer_1": "144", "answer_2": "124", "answer_3": "163"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p0": "0.200", "p3": "0.050", "p1": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, April 11, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 162 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "54", "answer_2": "85", "answer_3": "77", "answer_0": "72"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 6 grams of salt per serving and serves 5 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "13", "answer_0": "27", "answer_3": "21", "answer_2": "6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Thursday, May 21, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Saturday, January 03, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_3": "15", "answer_0": "33", "answer_1": "23"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "70", "answer_3": "42", "answer_2": "28", "answer_1": "34"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = closed; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 424:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 8 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "32", "answer_2": "24", "answer_3": "8", "answer_1": "26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1456 per year. It increases by 25% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1820", "answer_3": "1778", "answer_2": "1092", "answer_0": "1787"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=166, unit_price=539, discount_pct=0, region=EU, sku_total=705 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "8947", "v1": "705", "v3": "89308", "v0": "89474"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=623, unit_price=899, discount_pct=0, region=EU, sku_total=1522 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "560077", "v2": "56007", "v3": "560700", "v1": "1522"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p2": "0.083", "p1": "0.250", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $239. A fee of $62 is charged, then a deposit of $33 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "181", "answer_1": "184", "answer_3": "159", "answer_0": "179"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "23", "answer_2": "20", "answer_0": "32", "answer_3": "12"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = $100; priority = high.\n\nClaim to verify: \"the date is $100\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_1": "28", "answer_3": "32", "answer_2": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 8 grams of salt per serving and serves 11 people. A shopper bought a 88 gram box and a 880 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "20", "answer_3": "25", "answer_1": "25", "answer_0": "45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed February 27, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 11", "date_1": "March 09", "date_2": "March 15", "date_0": "March 13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.500", "p3": "0.167", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $841 per year. It increases by 5% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "798.95", "answer_0": "837.05", "answer_3": "808.05", "answer_1": "883.05"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = 2026-03-12; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $503 per year. It increases by 5% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "471.15", "answer_1": "528.15", "answer_0": "494.15", "answer_2": "477.85"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed May 13, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 25", "date_0": "May 27", "date_2": "May 25", "date_1": "May 23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=556, unit_price=918, discount_pct=0, region=EU, sku_total=1474 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "510964", "v2": "51040", "v0": "510408", "v1": "1474"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1179 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1120.05", "answer_1": "1237.95", "answer_0": "1205.95", "answer_3": "1231.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=672, unit_price=473, discount_pct=0, region=EU, sku_total=1145 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "317856", "v3": "318528", "v2": "31785", "v1": "1145"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $774. A fee of $80 is charged, then a deposit of $90 arrives. A second fee of $40 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "737", "answer_0": "744", "answer_3": "736", "answer_2": "757"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed January 03, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 12", "date_3": "January 14", "date_2": "January 13", "date_0": "January 15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 08:30 to 08:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "-1h45", "dur_0": "-1h30", "dur_1": "0h15", "dur_3": "-1h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.375", "p3": "0.125", "p1": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.125", "p0": "0.375", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = $682; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.500", "p0": "0.667", "p1": "0.833"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 06:00 to 12:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "6h30", "dur_3": "5h30", "dur_1": "6h30", "dur_0": "6h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 340 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "58", "answer_2": "57", "answer_3": "51", "answer_0": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1978. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "45", "age_2": "47", "age_0": "46", "age_3": "44"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=334, unit_price=703, discount_pct=0, region=EU, sku_total=1037 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "23480", "v1": "1037", "v3": "234468", "v0": "234802"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.167", "p0": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $22k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 246 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "203", "answer_3": "178", "answer_1": "196", "answer_0": "192"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 604:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=186, unit_price=101, discount_pct=0, region=EU, sku_total=287 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "1878", "v3": "18600", "v1": "287", "v0": "18786"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = 2026-06-19; priority = high.\n\nClaim to verify: \"the status is 2026-06-19\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p3": "0.050", "p1": "0.150", "p0": "0.100"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.429", "p1": "0.533", "p2": "0.533", "p0": "0.467"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 591:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed September 06, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 12", "date_2": "September 12", "date_3": "September 14", "date_0": "September 14"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed June 27, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 03", "date_2": "July 08", "date_0": "July 06", "date_3": "July 05"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Friday, February 27, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1969. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "54", "age_0": "56", "age_1": "55", "age_2": "57"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1986. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "40", "age_1": "39", "age_3": "42", "age_2": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1978. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "47", "age_3": "49", "age_1": "46", "age_2": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed January 20, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 26", "date_2": "January 28", "date_3": "January 26", "date_1": "January 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1983. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "42", "age_3": "39", "age_1": "40", "age_0": "41"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = 2026-05-06; priority = low.\n\nClaim to verify: \"the date is 2026-05-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1982. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "42", "age_3": "45", "age_2": "44", "age_0": "43"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed October 06, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 12", "date_0": "October 14", "date_3": "October 14", "date_2": "October 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = $738; priority = low.\n\nClaim to verify: \"the status is $738\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed April 25, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 02", "date_2": "May 07", "date_3": "May 04", "date_0": "May 05"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.500", "p3": "0.167", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.538", "p3": "0.417", "p1": "0.538", "p0": "0.462"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $389. A fee of $41 is charged, then a deposit of $46 arrives. A second fee of $20 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "365", "answer_2": "389", "answer_0": "374", "answer_3": "345"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 309 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "61", "answer_1": "63", "answer_0": "48", "answer_3": "29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 08:30 to 14:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "6h00", "dur_3": "5h00", "dur_2": "6h00", "dur_0": "5h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1975. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "49", "age_2": "51", "age_0": "50", "age_3": "52"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1524 per year. It increases by 5% this year, then a $40 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1571.2", "answer_2": "1447.8", "answer_1": "1600.2", "answer_0": "1560.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1962. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "64", "age_0": "63", "age_3": "65", "age_1": "62"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.083", "p1": "0.417", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = $329; priority = high.\n\nClaim to verify: \"the owner is $329\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed June 23, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 30", "date_1": "June 28", "date_0": "June 30", "date_2": "June 28"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 7 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $354 per year. It increases by 20% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "359.8", "answer_1": "424.8", "answer_0": "387.8", "answer_2": "283.2"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $848. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed January 14, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 26", "date_1": "January 24", "date_2": "January 26", "date_0": "January 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $285. A fee of $156 is charged, then a deposit of $41 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "92", "answer_2": "73", "answer_1": "119", "answer_3": "71"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $207. A fee of $51 is charged, then a deposit of $28 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "152", "answer_1": "173", "answer_0": "159", "answer_3": "138"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, July 09, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = ops; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1115 per year. It increases by 20% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1318", "answer_3": "1291", "answer_1": "1338", "answer_2": "892"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = pending; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $740. A fee of $61 is charged, then a deposit of $20 arrives. A second fee of $30 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "662", "answer_3": "671", "answer_2": "663", "answer_0": "669"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = $497; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 73 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "216", "answer_3": "204", "answer_1": "202", "answer_2": "198"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.167", "p1": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.200", "p2": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 181 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "336", "answer_2": "326", "answer_1": "351", "answer_3": "345"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = 2026-05-15; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Thursday, December 10, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.583", "p1": "0.417", "p2": "0.417", "p3": "0.545"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1979. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "47", "age_2": "48", "age_1": "46", "age_3": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 16:30 to 18:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h15", "dur_0": "0h45", "dur_3": "0h30", "dur_1": "1h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed March 21, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 28", "date_1": "March 26", "date_2": "March 29", "date_0": "March 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 13:00 to 17:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "4h30", "dur_0": "4h15", "dur_2": "4h45", "dur_3": "3h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 655:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=722, unit_price=861, discount_pct=0, region=EU, sku_total=1583 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "621642", "v1": "1583", "v3": "620920", "v2": "62164"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 390 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "89", "answer_1": "104", "answer_0": "84", "answer_2": "95"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.167", "p2": "0.500", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Wednesday, February 25, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1981. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "44", "age_0": "43", "age_3": "41", "age_1": "42"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = $467; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 531:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 274 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "83", "answer_0": "90", "answer_3": "71", "answer_1": "104"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 457:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = 2026-09-11; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1270 per year. It increases by 5% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1333.5", "answer_2": "1206.5", "answer_0": "1290.5", "answer_3": "1314.5"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.571", "p2": "0.375", "p0": "0.625", "p1": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "31", "answer_3": "56", "answer_0": "80", "answer_2": "24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $31k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1922 per year. It increases by 5% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2018.1", "answer_3": "1940.1", "answer_2": "1825.9", "answer_0": "1980.1"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Wednesday, January 07, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "25", "answer_1": "15", "answer_0": "35", "answer_2": "10"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $25k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.333", "p1": "0.417", "p3": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = $332; priority = low.\n\nClaim to verify: \"the status is $332\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 901:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $18k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.364", "p2": "0.364", "p3": "0.600", "p0": "0.636"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $381. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = $456; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed November 14, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 19", "date_0": "November 20", "date_3": "November 21", "date_2": "November 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 807:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = $765; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1550 per year. It increases by 25% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1891.5", "answer_1": "1937.5", "answer_2": "1162.5", "answer_3": "1877.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $331 per year. It increases by 15% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "345.65", "answer_1": "380.65", "answer_0": "353.65", "answer_2": "281.35"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_0": "36", "answer_3": "12", "answer_1": "26"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 12:00 to 21:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "8h30", "dur_2": "8h45", "dur_3": "8h00", "dur_1": "9h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $868. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=341, unit_price=296, discount_pct=0, region=EU, sku_total=637 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "100936", "v2": "10093", "v1": "637", "v3": "100595"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p1": "0.250", "p0": "0.200", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = $355; priority = low.\n\nClaim to verify: \"the amount is $355\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = open; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1338 per year. It increases by 25% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1672.5", "answer_0": "1639.5", "answer_3": "1651.5", "answer_2": "1003.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1959. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "64", "age_0": "65", "age_2": "66", "age_3": "63"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 6 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $827. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1452 per year. It increases by 20% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1730.4", "answer_1": "1742.4", "answer_2": "1161.6", "answer_3": "1769.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.286", "p1": "0.714", "p2": "0.714", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1546 per year. It increases by 10% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1644.6", "answer_0": "1657.6", "answer_2": "1391.4", "answer_1": "1700.6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=609, unit_price=289, discount_pct=0, region=EU, sku_total=898 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "898", "v0": "176001", "v3": "176610", "v2": "17600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.455", "p0": "0.500", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 758:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "12", "answer_2": "9", "answer_3": "9", "answer_0": "18"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Thursday, January 08, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.333", "p1": "0.417", "p3": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $26k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 236:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = 2026-08-05; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed November 22, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 26", "date_1": "November 26", "date_2": "November 28", "date_3": "November 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $26k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 886:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $3k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Friday, January 02, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $650. A fee of $95 is charged, then a deposit of $86 arrives. A second fee of $47 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "594", "answer_3": "621", "answer_2": "624", "answer_1": "599"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $849. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 386 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_1": "67", "answer_0": "50", "answer_2": "64"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 16:45 to 18:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h30", "dur_3": "0h15", "dur_1": "1h15", "dur_0": "1h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=135, unit_price=107, discount_pct=0, region=EU, sku_total=242 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "242", "v3": "14580", "v2": "1444", "v0": "14445"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 221 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "26", "answer_0": "36", "answer_1": "46", "answer_2": "30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 756:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = 2026-01-14; priority = low.\n\nClaim to verify: \"the owner is 2026-01-14\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed July 09, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 19", "date_3": "July 21", "date_0": "July 23", "date_2": "July 21"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $680. A fee of $43 is charged, then a deposit of $32 arrives. A second fee of $21 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "631", "answer_0": "648", "answer_1": "678", "answer_2": "661"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_2": "18", "answer_0": "30", "answer_1": "20"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, March 18, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 10 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = 2026-07-15; priority = low.\n\nClaim to verify: \"the status is 2026-07-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.083", "p2": "0.250", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=298, unit_price=312, discount_pct=0, region=EU, sku_total=610 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "92976", "v2": "9297", "v1": "610", "v3": "92678"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = ops; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.333", "p3": "0.600", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 910:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 07:00 to 09:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h00", "dur_1": "2h30", "dur_2": "2h30", "dur_3": "1h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = 2026-05-25; priority = high.\n\nClaim to verify: \"the amount is 2026-05-25\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.727", "p0": "0.273", "p3": "0.200", "p1": "0.727"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $506 per year. It increases by 15% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "531.9", "answer_3": "558.9", "answer_1": "581.9", "answer_2": "430.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 99 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_2": "69", "answer_3": "75", "answer_1": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_1": "26", "answer_3": "16", "answer_2": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 76 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "115", "answer_2": "92", "answer_0": "98", "answer_3": "84"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $510. A fee of $167 is charged, then a deposit of $59 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "328", "answer_3": "325", "answer_1": "290", "answer_0": "319"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed June 26, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 04", "date_0": "July 02", "date_1": "June 30", "date_3": "July 02"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 282 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "85", "answer_0": "96", "answer_1": "102", "answer_2": "112"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 779:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1891 per year. It increases by 10% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2080.1", "answer_0": "2059.1", "answer_2": "1701.9", "answer_3": "2045.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=424, unit_price=668, discount_pct=0, region=EU, sku_total=1092 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1092", "v0": "283232", "v2": "28323", "v3": "283656"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, June 18, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed April 17, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 21", "date_3": "April 23", "date_2": "April 25", "date_0": "April 23"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, June 28, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = $273; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $651. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = 2026-02-24; priority = low.\n\nClaim to verify: \"the status is 2026-02-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 201 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "29", "answer_1": "25", "answer_0": "40", "answer_3": "29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1998 per year. It increases by 10% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2197.8", "answer_3": "2147.8", "answer_2": "1798.2", "answer_0": "2180.8"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $773. A fee of $93 is charged, then a deposit of $78 arrives. A second fee of $46 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "702", "answer_3": "728", "answer_2": "701", "answer_0": "712"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $493 per year. It increases by 15% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "523.95", "answer_2": "419.05", "answer_1": "566.95", "answer_3": "484.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $15k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=178, unit_price=313, discount_pct=0, region=EU, sku_total=491 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "491", "v0": "55714", "v3": "55892", "v2": "5571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed March 09, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 15", "date_2": "March 19", "date_0": "March 17", "date_3": "March 17"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=342, unit_price=653, discount_pct=0, region=EU, sku_total=995 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "995", "v3": "222984", "v2": "22332", "v0": "223326"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=675, unit_price=637, discount_pct=0, region=EU, sku_total=1312 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "430650", "v2": "42997", "v0": "429975", "v1": "1312"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 541:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = $988; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 13:45 to 18:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h45", "dur_0": "3h30", "dur_3": "3h15", "dur_1": "4h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 12 people. A shopper bought a 84 gram box and a 840 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $254. A fee of $168 is charged, then a deposit of $56 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "77", "answer_0": "58", "answer_1": "28", "answer_3": "44"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.583", "p0": "0.417", "p3": "0.364", "p1": "0.583"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p1": "0.600", "p3": "0.250", "p2": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.385", "p0": "0.615", "p1": "0.385", "p3": "0.583"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.167", "p3": "0.083", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "15", "answer_0": "40", "answer_3": "25", "answer_1": "20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed January 25, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 04", "date_3": "February 06", "date_0": "February 06", "date_2": "February 04"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = 2026-05-08; priority = low.\n\nClaim to verify: \"the date is 2026-05-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=396, unit_price=810, discount_pct=0, region=EU, sku_total=1206 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "32076", "v1": "1206", "v3": "320364", "v0": "320760"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, July 23, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.364", "p3": "0.300", "p2": "0.636", "p1": "0.636"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed April 14, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 23", "date_2": "April 25", "date_3": "April 25", "date_0": "April 27"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p2": "0.100", "p3": "0.050", "p0": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=254, unit_price=187, discount_pct=0, region=EU, sku_total=441 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "441", "v0": "47498", "v3": "47752", "v2": "4749"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $116. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.583", "p1": "0.385", "p0": "0.615", "p2": "0.385"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Sunday, November 15, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=549, unit_price=530, discount_pct=0, region=EU, sku_total=1079 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "290421", "v1": "1079", "v2": "29097", "v0": "290970"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $26k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 09:15 to 17:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "7h30", "dur_1": "8h30", "dur_0": "8h00", "dur_2": "8h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "44", "answer_2": "24", "answer_3": "20", "answer_1": "29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 438:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $513. A fee of $54 is charged, then a deposit of $72 arrives. A second fee of $27 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "480", "answer_1": "514", "answer_0": "504", "answer_3": "491"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 07:30 to 22:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "15h00", "dur_0": "14h30", "dur_1": "15h15", "dur_3": "14h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $510. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.125", "p0": "0.250", "p1": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = pending; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p2": "0.500", "p3": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 5 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 17:45 to 19:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h00", "dur_0": "0h45", "dur_1": "1h30", "dur_3": "0h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 5 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 336 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "44", "answer_0": "64", "answer_1": "44", "answer_3": "49"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 7 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.273", "p1": "0.667", "p2": "0.667", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, December 26, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = 2026-11-27; priority = low.\n\nClaim to verify: \"the date is 2026-11-27\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1978. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "49", "age_1": "47", "age_3": "46", "age_0": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 181 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "84", "answer_3": "88", "answer_1": "80", "answer_2": "82"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=618, unit_price=852, discount_pct=0, region=EU, sku_total=1470 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "525918", "v0": "526536", "v1": "1470", "v2": "52653"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = $738; priority = high.\n\nClaim to verify: \"the amount is $738\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 389 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "138", "answer_1": "116", "answer_0": "125", "answer_2": "110"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = $258; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=557, unit_price=262, discount_pct=0, region=EU, sku_total=819 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "819", "v2": "14593", "v0": "145934", "v3": "146491"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 9 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = 2026-02-14; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 6 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "27", "answer_0": "63", "answer_3": "42", "answer_2": "21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.400", "p2": "0.400", "p3": "0.556", "p0": "0.600"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 20:30 to 20:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h15", "dur_1": "0h15", "dur_0": "-1h45", "dur_2": "0h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.455", "p1": "0.455", "p0": "0.545", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $432. A fee of $174 is charged, then a deposit of $74 arrives. A second fee of $87 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "245", "answer_1": "256", "answer_3": "249", "answer_2": "226"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1967. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_1": "57", "age_3": "60", "age_2": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Friday, December 18, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, January 20, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.333", "p3": "0.625", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.625", "p1": "0.333", "p0": "0.667", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An account starts at $851. A fee of $30 is charged, then a deposit of $50 arrives. A second fee of $15 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "856", "answer_3": "827", "answer_1": "847", "answer_2": "850"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Friday, October 02, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.167", "p0": "0.250", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.600", "p2": "0.400", "p3": "0.500", "p1": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.222", "p1": "0.700", "p2": "0.700", "p0": "0.300"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.333", "p2": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed September 09, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 15", "date_3": "September 15", "date_2": "September 13", "date_1": "September 13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, October 24, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "19", "answer_2": "16", "answer_0": "28", "answer_3": "12"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 03, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 16", "date_1": "April 12", "date_3": "April 14", "date_2": "April 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Tuesday, June 09, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "18", "answer_0": "24", "answer_2": "16", "answer_3": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1018 per year. It increases by 10% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1119.8", "answer_2": "916.2", "answer_3": "1131.8", "answer_0": "1105.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.667", "p3": "0.625", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $2k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed October 08, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 15", "date_3": "October 17", "date_0": "October 19", "date_2": "October 21"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, March 01, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 15:45 to 19:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h45", "dur_1": "3h15", "dur_2": "3h00", "dur_3": "2h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 72 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_1": "22", "answer_2": "28", "answer_3": "29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p3": "0.125", "p1": "0.500", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 812:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.300", "p2": "0.636", "p1": "0.636", "p0": "0.364"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 261:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.200", "p0": "0.800", "p3": "0.778", "p1": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 664:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "22", "answer_0": "28", "answer_2": "20", "answer_3": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.385", "p1": "0.385", "p3": "0.583", "p0": "0.615"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.500", "p0": "0.500", "p3": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = 2026-03-23; priority = low.\n\nClaim to verify: \"the date is 2026-03-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed May 18, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 26", "date_3": "May 28", "date_1": "May 26", "date_0": "May 28"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $611. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 10 people. A shopper bought a 70 gram box and a 700 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 8 grams of salt per serving and serves 7 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $553. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $966 per year. It increases by 20% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1159.2", "answer_2": "772.8", "answer_0": "1146.2", "answer_3": "1176.2"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, October 17, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=130, unit_price=574, discount_pct=0, region=EU, sku_total=704 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "7462", "v0": "74620", "v1": "704", "v3": "74750"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.455", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 5 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: date = 2026-12-11; priority = high.\n\nClaim to verify: \"the date is 2026-12-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=574, unit_price=766, discount_pct=0, region=EU, sku_total=1340 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1340", "v2": "43968", "v3": "440258", "v0": "439684"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p0": "0.150", "p1": "0.200", "p3": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1554 per year. It increases by 5% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1602.7", "answer_2": "1476.3", "answer_1": "1631.7", "answer_0": "1589.7"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed March 25, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 02", "date_2": "March 31", "date_0": "April 02", "date_1": "March 31"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 20:45 to 21:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "0h30", "dur_2": "0h30", "dur_0": "0h15", "dur_3": "-1h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.250", "p1": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.125", "p1": "0.375", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = $295; priority = high.\n\nClaim to verify: \"the amount is $295\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "63", "age_2": "65", "age_0": "64", "age_3": "66"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed October 03, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 09", "date_3": "October 11", "date_2": "October 10", "date_0": "October 12"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=236, unit_price=204, discount_pct=0, region=EU, sku_total=440 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "48144", "v1": "440", "v2": "4814", "v3": "48380"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $20k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1966. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_3": "56", "age_2": "59", "age_1": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $3k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = $746; priority = high.\n\nClaim to verify: \"the owner is $746\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $4k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1970. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "56", "age_0": "55", "age_1": "54", "age_3": "57"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.333", "p1": "0.500", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 11:30 to 22:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "11h00", "dur_1": "11h15", "dur_3": "10h15", "dur_0": "10h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 8 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 220 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "306", "answer_1": "338", "answer_2": "301", "answer_0": "320"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = 2026-07-13; priority = high.\n\nClaim to verify: \"the date is 2026-07-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 392 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "26", "answer_1": "48", "answer_0": "28", "answer_3": "31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p1": "0.571", "p0": "0.429", "p2": "0.571"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p2": "0.050", "p0": "0.100", "p3": "0.050"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 10:45 to 13:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h15", "dur_3": "2h00", "dur_1": "3h00", "dur_2": "2h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.583", "p0": "0.615", "p2": "0.385", "p1": "0.385"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1105 per year. It increases by 20% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1295", "answer_0": "1309", "answer_1": "1326", "answer_2": "884"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = ops; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = 2026-10-10; priority = low.\n\nClaim to verify: \"the owner is 2026-10-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_2": "12", "answer_1": "16", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1968. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "58", "age_3": "59", "age_1": "56", "age_0": "57"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.500", "p2": "0.167", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 706:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1475 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1401.25", "answer_0": "1516.75", "answer_1": "1548.75", "answer_3": "1493.75"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = 2026-07-05; priority = low.\n\nClaim to verify: \"the status is 2026-07-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, February 01, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=415, unit_price=237, discount_pct=0, region=EU, sku_total=652 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "9835", "v1": "652", "v3": "97940", "v0": "98355"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, June 12, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 11 people. A shopper bought a 66 gram box and a 660 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $989 per year. It increases by 20% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "791.2", "answer_1": "1186.8", "answer_3": "1187.8", "answer_0": "1163.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p3": "0.167", "p2": "0.333", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=841, unit_price=526, discount_pct=0, region=EU, sku_total=1367 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "44236", "v0": "442366", "v3": "443207", "v1": "1367"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $331. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = $377; priority = low.\n\nClaim to verify: \"the status is $377\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = pending; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $869. A fee of $129 is charged, then a deposit of $70 arrives. A second fee of $64 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "722", "answer_2": "744", "answer_0": "746", "answer_1": "771"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 11 people. A shopper bought a 99 gram box and a 990 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 8 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 258 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "124", "answer_3": "110", "answer_1": "110", "answer_0": "128"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $397. A fee of $81 is charged, then a deposit of $41 arrives. A second fee of $40 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "336", "answer_1": "327", "answer_0": "317", "answer_3": "344"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $992 per year. It increases by 25% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1203", "answer_2": "744", "answer_3": "1238", "answer_1": "1240"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Friday, October 16, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Thursday, June 04, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 382 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "336", "answer_2": "328", "answer_1": "329", "answer_3": "338"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 2 grams of salt per serving and serves 9 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1986. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "36", "age_1": "37", "age_2": "39", "age_0": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.800", "p3": "0.778", "p1": "0.200", "p2": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 170 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "37", "answer_1": "61", "answer_0": "42", "answer_2": "33"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = $274; priority = low.\n\nClaim to verify: \"the status is $274\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=589, unit_price=623, discount_pct=0, region=EU, sku_total=1212 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1212", "v0": "366947", "v2": "36694", "v3": "367536"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 483:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed November 03, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 09", "date_2": "November 13", "date_0": "November 11", "date_3": "November 11"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $698 per year. It increases by 20% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "841.6", "answer_2": "558.4", "answer_0": "817.6", "answer_1": "837.6"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed May 18, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 25", "date_0": "May 27", "date_2": "May 25", "date_3": "May 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 507:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = 2026-07-25; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $38k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p3": "0.333", "p1": "0.600", "p2": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = closed; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_0": "24", "answer_1": "12", "answer_2": "6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 10 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "45", "answer_2": "25", "answer_3": "20", "answer_1": "29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = $457; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 11 people. A shopper bought a 88 gram box and a 880 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = 2026-01-25; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.167", "p3": "0.083", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = open; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $684 per year. It increases by 25% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "806", "answer_0": "845", "answer_1": "855", "answer_2": "513"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $315. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 386 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "84", "answer_0": "75", "answer_2": "65", "answer_1": "63"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = $937; priority = high.\n\nClaim to verify: \"the date is $937\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 195 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "20", "answer_2": "43", "answer_1": "14", "answer_0": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 312 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "74", "answer_1": "84", "answer_0": "72", "answer_3": "87"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.385", "p2": "0.615", "p3": "0.333", "p1": "0.615"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_0": "64", "answer_1": "43", "answer_2": "40"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=745, unit_price=169, discount_pct=0, region=EU, sku_total=914 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "125905", "v1": "914", "v2": "12590", "v3": "125160"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1979. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "49", "age_2": "48", "age_1": "46", "age_0": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 11 people. A shopper bought a 66 gram box and a 660 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, July 28, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.429", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.500", "p2": "0.167", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1982. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "42", "age_2": "44", "age_3": "41", "age_0": "43"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = 2026-01-15; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.600", "p1": "0.400", "p3": "0.556", "p2": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=825, unit_price=904, discount_pct=0, region=EU, sku_total=1729 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "744975", "v0": "745800", "v1": "1729", "v2": "74580"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=426, unit_price=560, discount_pct=0, region=EU, sku_total=986 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "238986", "v0": "238560", "v2": "23856", "v1": "986"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Contract signed December 22, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 04", "date_1": "December 31", "date_2": "January 02", "date_3": "January 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed March 15, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 20", "date_3": "March 22", "date_2": "March 18", "date_0": "March 20"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed April 09, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 18", "date_0": "April 22", "date_2": "April 24", "date_3": "April 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1787 per year. It increases by 15% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1518.95", "answer_3": "2018.05", "answer_0": "2026.05", "answer_1": "2055.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=345, unit_price=650, discount_pct=0, region=EU, sku_total=995 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "22425", "v3": "223905", "v0": "224250", "v1": "995"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $498. A fee of $76 is charged, then a deposit of $35 arrives. A second fee of $38 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "401", "answer_0": "419", "answer_2": "422", "answer_1": "409"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 323:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.125", "p0": "0.375", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1988. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "38", "age_3": "35", "age_0": "37", "age_1": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.500", "p0": "0.500", "p3": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=616, unit_price=194, discount_pct=0, region=EU, sku_total=810 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "118888", "v2": "11950", "v0": "119504", "v1": "810"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Sunday, October 04, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $803 per year. It increases by 5% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "804.15", "answer_1": "843.15", "answer_0": "823.15", "answer_2": "762.85"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p3": "0.444", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1505 per year. It increases by 15% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1669.75", "answer_0": "1705.75", "answer_1": "1730.75", "answer_2": "1279.25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 260 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "82", "answer_3": "73", "answer_2": "65", "answer_0": "80"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $461 per year. It increases by 15% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "391.85", "answer_3": "455.15", "answer_0": "489.15", "answer_1": "530.15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 3 grams of salt per serving and serves 8 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 09:00 to 22:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "13h45", "dur_2": "13h45", "dur_3": "12h45", "dur_0": "13h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1959. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "64", "age_0": "65", "age_2": "66", "age_3": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.417", "p0": "0.333", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $361 per year. It increases by 10% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "351.1", "answer_3": "388.1", "answer_2": "324.9", "answer_1": "397.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "35", "answer_3": "25", "answer_1": "15", "answer_2": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1979. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "48", "age_3": "45", "age_0": "47", "age_1": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 130 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "92", "answer_1": "83", "answer_2": "63", "answer_0": "72"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1763 per year. It increases by 10% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1586.7", "answer_0": "1927.3", "answer_3": "1888.3", "answer_1": "1939.3"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 11 people. A shopper bought a 99 gram box and a 990 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $7k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $475 per year. It increases by 25% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "566.75", "answer_2": "356.25", "answer_3": "540.75", "answer_1": "593.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p0": "0.583", "p3": "0.545", "p2": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $93. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_3": "12", "answer_0": "36", "answer_1": "27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = $134; priority = high.\n\nClaim to verify: \"the date is $134\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $824 per year. It increases by 25% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1012", "answer_2": "618", "answer_1": "1030", "answer_3": "983"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 273 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "55", "answer_1": "53", "answer_0": "48", "answer_2": "51"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "29", "answer_2": "24", "answer_0": "54", "answer_3": "30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 117 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_2": "100", "answer_1": "62", "answer_3": "77"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1959. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "65", "age_0": "66", "age_2": "67", "age_3": "68"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 84 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "225", "answer_3": "243", "answer_0": "240", "answer_2": "234"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = pending; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Wednesday, April 01, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "25", "answer_3": "28", "answer_2": "21", "answer_0": "49"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, May 03, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed March 18, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 21", "date_3": "March 23", "date_2": "March 25", "date_0": "March 23"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $754. A fee of $72 is charged, then a deposit of $55 arrives. A second fee of $36 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "683", "answer_0": "701", "answer_3": "717", "answer_1": "715"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $275. A fee of $116 is charged, then a deposit of $42 arrives. A second fee of $58 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "126", "answer_2": "149", "answer_3": "119", "answer_0": "143"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, September 26, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1986 per year. It increases by 15% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2216.9", "answer_2": "1688.1", "answer_1": "2283.9", "answer_0": "2242.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 3 people. A shopper bought a 9 gram box and a 90 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1981. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "45", "age_2": "44", "age_1": "42", "age_0": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 07:00 to 11:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "4h30", "dur_0": "4h15", "dur_1": "4h30", "dur_3": "3h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = 2026-09-02; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, September 12, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=316, unit_price=526, discount_pct=0, region=EU, sku_total=842 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "166532", "v2": "16621", "v0": "166216", "v1": "842"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $708. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = $120; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1959. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "66", "age_2": "68", "age_3": "65", "age_0": "67"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = $197; priority = high.\n\nClaim to verify: \"the owner is $197\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 08:30 to 13:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "4h15", "dur_0": "4h00", "dur_1": "4h45", "dur_3": "3h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $808 per year. It increases by 15% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "686.8", "answer_0": "885.2", "answer_1": "929.2", "answer_3": "867.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.286", "p0": "0.375", "p2": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = $392; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "43", "answer_3": "24", "answer_0": "64", "answer_2": "40"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 791:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.455", "p3": "0.400", "p2": "0.545", "p1": "0.545"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Wednesday, May 13, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 361 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "55", "answer_0": "72", "answer_1": "70", "answer_3": "80"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 291 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "112", "answer_2": "108", "answer_1": "127", "answer_3": "119"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $3k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.200", "p2": "0.667", "p1": "0.667", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = 2026-03-19; priority = high.\n\nClaim to verify: \"the owner is 2026-03-19\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1990. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "37", "age_0": "35", "age_1": "34", "age_2": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $5k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 876:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $899. A fee of $41 is charged, then a deposit of $27 arrives. A second fee of $20 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "882", "answer_2": "854", "answer_0": "865", "answer_1": "840"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = 2026-05-22; priority = low.\n\nClaim to verify: \"the owner is 2026-05-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $862. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.286", "p1": "0.714", "p3": "0.167", "p2": "0.714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.083", "p0": "0.333", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $823. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 211 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "135", "answer_1": "116", "answer_3": "125", "answer_2": "154"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = $934; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.250", "p2": "0.125", "p1": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p1": "0.375", "p3": "0.125", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = 2026-01-07; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = $820; priority = high.\n\nClaim to verify: \"the date is $820\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1975. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "49", "age_2": "52", "age_1": "50", "age_0": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $881. A fee of $61 is charged, then a deposit of $66 arrives. A second fee of $30 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "876", "answer_0": "856", "answer_2": "834", "answer_3": "832"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = 2026-11-08; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.444", "p1": "0.556", "p3": "0.375", "p2": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.083", "p0": "0.333", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 203:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 384:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, December 28, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = $236; priority = low.\n\nClaim to verify: \"the owner is $236\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $881. A fee of $124 is charged, then a deposit of $82 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "777", "answer_3": "768", "answer_1": "762", "answer_2": "792"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed May 06, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 16", "date_2": "May 22", "date_0": "May 20", "date_3": "May 18"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = $848; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $596. A fee of $37 is charged, then a deposit of $39 arrives. A second fee of $18 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "559", "answer_0": "580", "answer_1": "605", "answer_3": "555"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $214. A fee of $152 is charged, then a deposit of $36 arrives. A second fee of $76 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "22", "answer_3": "35", "answer_1": "51", "answer_2": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p1": "0.833", "p2": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.667", "p1": "0.833", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $58. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 764:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, May 28, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Saturday, January 17, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed September 06, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 17", "date_2": "September 15", "date_3": "September 17", "date_1": "September 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1982. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "41", "age_2": "43", "age_0": "42", "age_3": "40"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 53 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "70", "answer_3": "41", "answer_1": "46", "answer_0": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = $656; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1972. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "52", "age_2": "54", "age_0": "53", "age_3": "55"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $288. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.083", "p0": "0.250", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.250", "p0": "0.333", "p1": "0.417"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed May 26, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 05", "date_2": "June 11", "date_3": "June 07", "date_0": "June 09"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=987, unit_price=741, discount_pct=0, region=EU, sku_total=1728 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "731367", "v3": "732354", "v1": "1728", "v2": "73136"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 293 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "62", "answer_0": "45", "answer_1": "52", "answer_2": "64"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.444", "p2": "0.444", "p0": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 55 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_1": "41", "answer_2": "45", "answer_3": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 9 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "28", "answer_2": "20", "answer_1": "22", "answer_3": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed June 21, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 02", "date_1": "June 30", "date_2": "July 04", "date_3": "July 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $40k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=372, unit_price=424, discount_pct=0, region=EU, sku_total=796 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "157728", "v2": "15772", "v3": "158100", "v1": "796"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 8 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $680. A fee of $93 is charged, then a deposit of $76 arrives. A second fee of $46 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "596", "answer_0": "617", "answer_2": "644", "answer_3": "609"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1958. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "69", "age_0": "68", "age_1": "67", "age_3": "66"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed July 13, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 21", "date_3": "July 23", "date_2": "July 25", "date_0": "July 23"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.538", "p0": "0.462", "p2": "0.538", "p3": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = $841; priority = low.\n\nClaim to verify: \"the status is $841\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.667", "p2": "0.667", "p3": "0.273"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $515 per year. It increases by 15% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "563.25", "answer_3": "558.25", "answer_2": "437.75", "answer_1": "592.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = $740; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = $333; priority = low.\n\nClaim to verify: \"the status is $333\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $248 per year. It increases by 10% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "197.8", "answer_2": "223.2", "answer_0": "229.8", "answer_1": "272.8"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $701. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "29", "answer_2": "24", "answer_3": "40", "answer_0": "64"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.571", "p0": "0.625", "p2": "0.375", "p1": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed July 08, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 18", "date_3": "July 20", "date_0": "July 22", "date_2": "July 20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, September 05, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 9 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed August 21, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "August 25", "date_1": "August 25", "date_3": "August 27", "date_0": "August 27"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = $313; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: status = ops; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = $315; priority = high.\n\nClaim to verify: \"the status is $315\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.333", "p0": "0.250", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed October 26, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 03", "date_0": "November 05", "date_3": "November 05", "date_1": "November 03"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_0": "45", "answer_3": "15", "answer_1": "33"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = $972; priority = high.\n\nClaim to verify: \"the amount is $972\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: status = open; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, February 20, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = $423; priority = low.\n\nClaim to verify: \"the date is $423\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.462", "p1": "0.500", "p0": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 3 grams of salt per serving and serves 8 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p3": "0.050", "p2": "0.050", "p1": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 524:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=703, unit_price=177, discount_pct=0, region=EU, sku_total=880 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "12443", "v0": "124431", "v1": "880", "v3": "123728"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=412, unit_price=102, discount_pct=0, region=EU, sku_total=514 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "4202", "v1": "514", "v3": "41612", "v0": "42024"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.778", "p3": "0.125", "p0": "0.222", "p1": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_1": "31", "answer_3": "56", "answer_0": "80"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 210:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $345. A fee of $95 is charged, then a deposit of $43 arrives. A second fee of $47 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "246", "answer_2": "237", "answer_1": "272", "answer_3": "234"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed July 04, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 13", "date_3": "July 15", "date_0": "July 16", "date_2": "July 14"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 249:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $21k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 463:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = $324; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = 2026-08-24; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Monday, April 13, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, July 21, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1186 per year. It increases by 20% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1405.2", "answer_3": "1392.2", "answer_1": "1423.2", "answer_2": "948.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.500", "p3": "0.167", "p1": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.417", "p0": "0.333", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.583", "p0": "0.615", "p1": "0.385", "p2": "0.385"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.667", "p1": "0.333", "p3": "0.636"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 177 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "88", "answer_2": "101", "answer_0": "84", "answer_3": "92"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1966. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "58", "age_3": "61", "age_2": "60", "age_0": "59"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = $678; priority = high.\n\nClaim to verify: \"the amount is $678\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 563:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 355 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "112", "answer_3": "109", "answer_0": "128", "answer_1": "109"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=979, unit_price=726, discount_pct=0, region=EU, sku_total=1705 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "710754", "v2": "71075", "v1": "1705", "v3": "711733"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=571, unit_price=295, discount_pct=0, region=EU, sku_total=866 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "167874", "v1": "866", "v0": "168445", "v2": "16844"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.167", "p1": "0.500", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1988. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "35", "age_0": "36", "age_3": "34", "age_2": "37"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.200", "p2": "0.727", "p1": "0.727", "p0": "0.273"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $29k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.333", "p0": "0.667", "p3": "0.600"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.750", "p3": "0.714", "p1": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_2": "28", "answer_1": "30", "answer_0": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = 2026-01-18; priority = high.\n\nClaim to verify: \"the status is 2026-01-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 334 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "160", "answer_2": "147", "answer_3": "146", "answer_0": "144"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, November 16, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $704. A fee of $160 is charged, then a deposit of $58 arrives. A second fee of $80 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "531", "answer_0": "522", "answer_3": "516", "answer_2": "519"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 148 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "128", "answer_2": "118", "answer_1": "112", "answer_0": "120"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = $774; priority = low.\n\nClaim to verify: \"the status is $774\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "12", "answer_0": "18", "answer_3": "9", "answer_2": "9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 492:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.500", "p3": "0.167", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 5 grams of salt per serving and serves 6 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 485:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.385", "p2": "0.571", "p0": "0.429", "p1": "0.571"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p1": "0.800", "p2": "0.800", "p3": "0.111"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_1": "22", "answer_0": "64", "answer_3": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1994 per year. It increases by 15% this year, then a $40 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1694.9", "answer_3": "2274.1", "answer_1": "2293.1", "answer_0": "2253.1"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $786. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $743 per year. It increases by 5% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "803.15", "answer_2": "705.85", "answer_1": "780.15", "answer_0": "770.15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed December 01, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 06", "date_0": "December 04", "date_2": "December 06", "date_1": "December 04"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1019 per year. It increases by 25% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1250.75", "answer_3": "1275.75", "answer_1": "1273.75", "answer_2": "764.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 215:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.778", "p3": "0.125", "p0": "0.222", "p1": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, January 25, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 13:30 to 14:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "0h00", "dur_2": "1h00", "dur_1": "1h00", "dur_0": "0h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 59 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "220", "answer_1": "227", "answer_3": "201", "answer_0": "216"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $899. A fee of $82 is charged, then a deposit of $75 arrives. A second fee of $41 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "851", "answer_1": "827", "answer_2": "858", "answer_3": "840"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $699. A fee of $157 is charged, then a deposit of $66 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "554", "answer_0": "530", "answer_2": "525", "answer_3": "523"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed June 13, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 22", "date_1": "June 20", "date_0": "June 23", "date_2": "June 25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.600", "p3": "0.250", "p1": "0.600", "p0": "0.400"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, February 12, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 9 grams of salt per serving and serves 10 people. A shopper bought a 90 gram box and a 900 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1972. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "52", "age_3": "55", "age_0": "53", "age_2": "54"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.429", "p0": "0.571", "p3": "0.538", "p2": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=745, unit_price=103, discount_pct=0, region=EU, sku_total=848 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "76735", "v3": "77480", "v1": "848", "v2": "7673"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.429", "p0": "0.571", "p2": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_2": "24", "answer_1": "28", "answer_0": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=435, unit_price=813, discount_pct=0, region=EU, sku_total=1248 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "35365", "v1": "1248", "v3": "354090", "v0": "353655"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = 2026-06-23; priority = low.\n\nClaim to verify: \"the date is 2026-06-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=265, unit_price=850, discount_pct=0, region=EU, sku_total=1115 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1115", "v3": "224985", "v2": "22525", "v0": "225250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 285:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "41", "answer_3": "30", "answer_0": "66", "answer_2": "36"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = 2026-03-05; priority = high.\n\nClaim to verify: \"the date is 2026-03-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1982. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "46", "age_1": "43", "age_0": "44", "age_2": "45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = 2026-05-09; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = 2026-09-23; priority = low.\n\nClaim to verify: \"the owner is 2026-09-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "6", "answer_1": "8", "answer_0": "12", "answer_3": "6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1970. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "56", "age_1": "55", "age_2": "57", "age_3": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p2": "0.800", "p1": "0.800", "p3": "0.111"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=264, unit_price=990, discount_pct=0, region=EU, sku_total=1254 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "26136", "v3": "261624", "v1": "1254", "v0": "261360"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p2": "0.125", "p3": "0.125", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed July 28, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 05", "date_0": "August 07", "date_3": "August 07", "date_2": "August 09"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=211, unit_price=940, discount_pct=0, region=EU, sku_total=1151 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "198129", "v2": "19834", "v1": "1151", "v0": "198340"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = 2026-06-17; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "32", "answer_2": "40", "answer_0": "72", "answer_1": "44"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_3": "20", "answer_1": "24", "answer_2": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.333", "p3": "0.600", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = 2026-07-22; priority = high.\n\nClaim to verify: \"the owner is 2026-07-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = 2026-04-18; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.364", "p3": "0.600", "p1": "0.364", "p0": "0.636"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = $655; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=891, unit_price=399, discount_pct=0, region=EU, sku_total=1290 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "354618", "v1": "1290", "v0": "355509", "v2": "35550"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed January 21, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 29", "date_1": "January 27", "date_0": "January 29", "date_2": "January 31"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=465, unit_price=132, discount_pct=0, region=EU, sku_total=597 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "61380", "v1": "597", "v2": "6138", "v3": "60915"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed December 15, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 26", "date_0": "December 24", "date_3": "December 24", "date_1": "December 22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $4k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = $181; priority = high.\n\nClaim to verify: \"the owner is $181\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $583. A fee of $86 is charged, then a deposit of $26 arrives. A second fee of $43 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "480", "answer_1": "498", "answer_3": "496", "answer_2": "457"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=865, unit_price=607, discount_pct=0, region=EU, sku_total=1472 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "52505", "v3": "525920", "v0": "525055", "v1": "1472"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = $599; priority = low.\n\nClaim to verify: \"the owner is $599\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1257 per year. It increases by 15% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1445.55", "answer_0": "1410.55", "answer_2": "1068.45", "answer_3": "1395.55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = $766; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 9 grams of salt per serving and serves 7 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p3": "0.125", "p1": "0.375", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, May 22, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = 2026-07-24; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "50", "answer_1": "22", "answer_2": "15", "answer_3": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $331. A fee of $113 is charged, then a deposit of $89 arrives. A second fee of $56 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "221", "answer_3": "228", "answer_0": "251", "answer_2": "238"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Contract signed May 18, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 29", "date_0": "May 29", "date_1": "May 27", "date_2": "May 31"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "21", "answer_2": "14", "answer_0": "35", "answer_1": "17"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, December 01, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.667", "p0": "0.500", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 912:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = $619; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.714", "p2": "0.714", "p0": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.500", "p3": "0.467", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.333", "p3": "0.167", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.800", "p1": "0.800", "p3": "0.111", "p0": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1953 per year. It increases by 20% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2343.6", "answer_0": "2328.6", "answer_2": "1562.4", "answer_3": "2351.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.500", "p3": "0.167", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1981. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "46", "age_1": "44", "age_0": "45", "age_3": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 766:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.375", "p1": "0.500", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "11", "answer_3": "6", "answer_2": "9", "answer_0": "15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p3": "0.050", "p2": "0.100", "p0": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $258. A fee of $158 is charged, then a deposit of $26 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "39", "answer_2": "57", "answer_3": "26", "answer_0": "47"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 411:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1047 per year. It increases by 15% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1204.05", "answer_1": "1204.05", "answer_0": "1165.05", "answer_2": "889.95"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $404. A fee of $163 is charged, then a deposit of $84 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "252", "answer_2": "270", "answer_3": "264", "answer_0": "244"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = open; priority = high.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p3": "0.050", "p2": "0.100", "p0": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.083", "p2": "0.167", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p3": "0.167", "p0": "0.500", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed November 10, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 14", "date_3": "November 16", "date_2": "November 14", "date_0": "November 16"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = $896; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1983. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "42", "age_0": "43", "age_3": "45", "age_2": "44"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = 2026-01-14; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = $880; priority = high.\n\nClaim to verify: \"the owner is $880\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.333", "p0": "0.667", "p3": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.286", "p3": "0.667", "p2": "0.286", "p0": "0.714"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.333", "p1": "0.667", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=120, unit_price=109, discount_pct=0, region=EU, sku_total=229 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "229", "v3": "12960", "v0": "13080", "v2": "1308"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Contract signed February 04, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 11", "date_3": "February 13", "date_0": "February 13", "date_2": "February 15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1515 per year. It increases by 25% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1136.25", "answer_3": "1889.75", "answer_1": "1893.75", "answer_0": "1866.75"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 09:15 to 16:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "6h00", "dur_2": "6h30", "dur_0": "6h15", "dur_1": "7h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 942:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1980. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "48", "age_0": "46", "age_1": "45", "age_2": "47"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $784 per year. It increases by 20% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "627.2", "answer_1": "940.8", "answer_0": "922.8", "answer_3": "888.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = 2026-11-05; priority = low.\n\nClaim to verify: \"the status is 2026-11-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed February 20, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 24", "date_0": "February 26", "date_2": "February 24", "date_3": "February 26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.400", "p0": "0.600", "p1": "0.400"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1969. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "56", "age_1": "55", "age_2": "57", "age_3": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "63", "answer_3": "21", "answer_1": "45", "answer_2": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, July 19, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = 2026-10-02; priority = high.\n\nClaim to verify: \"the owner is 2026-10-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1978. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "48", "age_0": "47", "age_1": "46", "age_3": "45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed February 16, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 18", "date_0": "February 20", "date_3": "February 22", "date_1": "February 20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p0": "0.500", "p3": "0.400"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = $310; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.545", "p1": "0.455", "p3": "0.500", "p2": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = 2026-04-08; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.167", "p1": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 273 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "139", "answer_1": "145", "answer_0": "126", "answer_3": "123"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, December 10, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = 2026-10-15; priority = high.\n\nClaim to verify: \"the status is 2026-10-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=962, unit_price=445, discount_pct=0, region=EU, sku_total=1407 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "428090", "v1": "1407", "v3": "427128", "v2": "42809"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $328. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = open; priority = high.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 3 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 9 grams of salt per serving and serves 10 people. A shopper bought a 90 gram box and a 900 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p0": "0.500", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 797:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_3": "21", "answer_1": "22", "answer_2": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = $844; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = $483; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=817, unit_price=427, discount_pct=0, region=EU, sku_total=1244 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "349676", "v1": "1244", "v2": "34885", "v0": "348859"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p3": "0.250", "p0": "0.400", "p2": "0.600"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "24", "answer_3": "21", "answer_2": "21", "answer_0": "42"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed March 20, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 04", "date_1": "March 29", "date_0": "April 02", "date_3": "March 31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 10:15 to 20:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "9h15", "dur_0": "9h00", "dur_3": "8h45", "dur_1": "9h45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=476, unit_price=160, discount_pct=0, region=EU, sku_total=636 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "76160", "v3": "75684", "v1": "636", "v2": "7616"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.714", "p0": "0.286", "p2": "0.714"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 4 grams of salt per serving and serves 6 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1983. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "42", "age_0": "41", "age_3": "39", "age_1": "40"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = 2026-12-01; priority = low.\n\nClaim to verify: \"the owner is 2026-12-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_0": "54", "answer_3": "30", "answer_1": "29"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "45", "answer_3": "35", "answer_1": "17", "answer_2": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 6 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $19k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_2": "18", "answer_0": "60", "answer_1": "25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = $241; priority = high.\n\nClaim to verify: \"the owner is $241\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, May 21, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 510:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.467", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "19", "answer_2": "12", "answer_0": "40", "answer_3": "28"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "9", "answer_1": "14", "answer_0": "24", "answer_3": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 7 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 306 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "168", "answer_1": "163", "answer_2": "179", "answer_3": "148"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 816:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.333", "p3": "0.167", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 586:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 466:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed June 08, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 13", "date_1": "June 11", "date_0": "June 11", "date_3": "June 13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.667", "p0": "0.333", "p3": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p2": "0.375", "p0": "0.500", "p3": "0.125"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = 2026-05-01; priority = low.\n\nClaim to verify: \"the date is 2026-05-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 100 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "125", "answer_3": "134", "answer_1": "129", "answer_2": "123"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1978. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "46", "age_1": "45", "age_2": "47", "age_3": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = $200; priority = high.\n\nClaim to verify: \"the status is $200\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.364", "p1": "0.364", "p0": "0.636", "p3": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1130 per year. It increases by 5% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1073.5", "answer_0": "1176.5", "answer_3": "1145.5", "answer_1": "1186.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 537:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 15:45 to 21:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "5h15", "dur_0": "5h00", "dur_3": "4h15", "dur_1": "5h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = $881; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, June 16, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1518 per year. It increases by 25% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1861.5", "answer_1": "1897.5", "answer_2": "1138.5", "answer_3": "1872.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = 2026-10-12; priority = high.\n\nClaim to verify: \"the date is 2026-10-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $318. A fee of $100 is charged, then a deposit of $72 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "227", "answer_1": "261", "answer_2": "251", "answer_0": "240"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $552. A fee of $43 is charged, then a deposit of $90 arrives. A second fee of $21 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "575", "answer_3": "575", "answer_0": "578", "answer_2": "588"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p3": "0.467", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 313 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "238", "answer_3": "215", "answer_2": "240", "answer_0": "224"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=624, unit_price=386, discount_pct=0, region=EU, sku_total=1010 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1010", "v0": "240864", "v3": "241488", "v2": "24086"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = closed; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $14k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed November 03, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 11", "date_3": "November 11", "date_1": "November 09", "date_2": "November 13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = $724; priority = high.\n\nClaim to verify: \"the owner is $724\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $470. A fee of $44 is charged, then a deposit of $24 arrives. A second fee of $22 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "428", "answer_2": "449", "answer_3": "430", "answer_1": "430"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 642:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 783:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1981. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "41", "age_2": "44", "age_1": "42", "age_0": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=672, unit_price=257, discount_pct=0, region=EU, sku_total=929 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "172032", "v1": "929", "v2": "17270", "v0": "172704"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1991 per year. It increases by 20% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2339.2", "answer_2": "1592.8", "answer_3": "2325.2", "answer_1": "2389.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 924:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = $780; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "10", "answer_0": "16", "answer_2": "8", "answer_3": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed December 26, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 31", "date_0": "January 01", "date_2": "January 03", "date_3": "January 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1974. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "52", "age_2": "53", "age_1": "51", "age_3": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = $950; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1982. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "44", "age_3": "42", "age_1": "43", "age_2": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "8", "answer_1": "10", "answer_2": "8", "answer_0": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 685:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, January 17, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $645. A fee of $63 is charged, then a deposit of $43 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "609", "answer_0": "594", "answer_2": "580", "answer_1": "566"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $504 per year. It increases by 10% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "554.4", "answer_3": "501.4", "answer_2": "453.6", "answer_0": "515.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = $870; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = 2026-05-20; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1665 per year. It increases by 5% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1748.25", "answer_3": "1731.25", "answer_2": "1581.75", "answer_0": "1719.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $25k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p3": "0.467", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1964. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "59", "age_0": "60", "age_3": "62", "age_2": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 553:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.333", "p3": "0.600", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = 2026-02-07; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.545", "p2": "0.545", "p0": "0.455", "p3": "0.400"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = $261; priority = high.\n\nClaim to verify: \"the owner is $261\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $527. A fee of $162 is charged, then a deposit of $61 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "367", "answer_1": "369", "answer_2": "350", "answer_0": "345"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.050", "p1": "0.150", "p0": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = $995; priority = high.\n\nClaim to verify: \"the date is $995\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.455", "p3": "0.400", "p2": "0.545", "p1": "0.545"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 3 grams of salt per serving and serves 11 people. A shopper bought a 33 gram box and a 330 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.667", "p2": "0.333", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = pending; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_1": "21", "answer_2": "18", "answer_3": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $630. A fee of $155 is charged, then a deposit of $65 arrives. A second fee of $77 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "463", "answer_1": "483", "answer_2": "475", "answer_3": "457"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = $837; priority = high.\n\nClaim to verify: \"the owner is $837\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 11 people. A shopper bought a 22 gram box and a 220 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 126 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "270", "answer_1": "273", "answer_0": "288", "answer_3": "293"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1477 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1518.85", "answer_2": "1403.15", "answer_3": "1548.85", "answer_1": "1550.85"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = ops; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, October 16, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.429", "p2": "0.533", "p1": "0.533", "p0": "0.467"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A shift runs from 06:15 to 16:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "9h15", "dur_1": "10h15", "dur_2": "10h15", "dur_0": "9h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed May 28, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 04", "date_2": "June 06", "date_3": "June 06", "date_0": "June 08"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $355. A fee of $73 is charged, then a deposit of $71 arrives. A second fee of $36 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "313", "answer_1": "306", "answer_0": "317", "answer_2": "301"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = $356; priority = high.\n\nClaim to verify: \"the status is $356\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed June 28, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 07", "date_2": "July 09", "date_1": "July 05", "date_3": "July 07"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=541, unit_price=308, discount_pct=0, region=EU, sku_total=849 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "167169", "v2": "16662", "v1": "849", "v0": "166628"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 239 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "313", "answer_3": "331", "answer_2": "300", "answer_0": "320"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 12 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 316:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1890 per year. It increases by 5% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1795.5", "answer_3": "1932.5", "answer_0": "1948.5", "answer_1": "1984.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.286", "p3": "0.667", "p1": "0.286", "p0": "0.714"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.500", "p0": "0.667", "p1": "0.833"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 912:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 350 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "141", "answer_1": "147", "answer_0": "160", "answer_2": "175"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Contract signed July 24, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 29", "date_2": "July 31", "date_0": "July 29", "date_1": "July 27"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1963. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "63", "age_1": "60", "age_0": "61", "age_2": "62"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p2": "0.200", "p0": "0.800", "p3": "0.778"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 16:30 to 21:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "4h00", "dur_3": "3h30", "dur_2": "4h15", "dur_1": "4h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed May 22, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 30", "date_2": "June 01", "date_3": "June 01", "date_0": "June 03"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed July 15, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 24", "date_3": "July 26", "date_0": "July 28", "date_2": "July 30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 19:45 to 21:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "1h00", "dur_1": "1h30", "dur_3": "0h30", "dur_2": "1h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.667", "p3": "0.167", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.571", "p1": "0.571", "p0": "0.429", "p3": "0.385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 884:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = $727; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.333", "p2": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 8 grams of salt per serving and serves 3 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 790:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1979. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "45", "age_3": "48", "age_0": "46", "age_2": "47"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $27k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $297. A fee of $91 is charged, then a deposit of $83 arrives. A second fee of $45 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "215", "answer_2": "224", "answer_0": "244", "answer_3": "231"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = pending; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_3": "12", "answer_1": "27", "answer_0": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p2": "0.600", "p3": "0.333", "p0": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.083", "p0": "0.333", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=658, unit_price=969, discount_pct=0, region=EU, sku_total=1627 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1627", "v2": "63760", "v0": "637602", "v3": "638260"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1987. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "36", "age_2": "38", "age_0": "37", "age_3": "35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1971. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "53", "age_3": "56", "age_0": "54", "age_2": "55"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1989. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "36", "age_2": "37", "age_3": "38", "age_1": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = closed; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p2": "0.500", "p3": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $667. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=307, unit_price=482, discount_pct=0, region=EU, sku_total=789 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "789", "v0": "147974", "v3": "148281", "v2": "14797"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $8k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p3": "0.455", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.667", "p3": "0.250", "p2": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p3": "0.050", "p2": "0.150", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1362 per year. It increases by 15% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1157.7", "answer_1": "1566.3", "answer_0": "1550.3", "answer_3": "1572.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_1": "38", "answer_2": "36", "answer_3": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = $886; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 84 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "104", "answer_0": "100", "answer_2": "83", "answer_1": "103"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 235 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "46", "answer_1": "45", "answer_3": "80", "answer_0": "64"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = $962; priority = low.\n\nClaim to verify: \"the date is $962\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $847 per year. It increases by 25% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1024.75", "answer_2": "635.25", "answer_0": "1015.75", "answer_1": "1058.75"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 141 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "59", "answer_1": "36", "answer_2": "44", "answer_0": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.545", "p0": "0.455", "p1": "0.545", "p3": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.200", "p2": "0.150", "p1": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.467", "p1": "0.533", "p3": "0.429", "p2": "0.533"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1968. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "59", "age_3": "60", "age_1": "57", "age_0": "58"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 738:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1283 per year. It increases by 5% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1218.85", "answer_1": "1347.15", "answer_0": "1327.15", "answer_3": "1338.15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = 2026-11-09; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.333", "p1": "0.333", "p3": "0.636"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 248 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "186", "answer_2": "197", "answer_1": "201", "answer_0": "192"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, August 25, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed May 02, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 06", "date_3": "May 07", "date_2": "May 04", "date_1": "May 05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.600", "p1": "0.400", "p2": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.167", "p1": "0.667", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $294. A fee of $68 is charged, then a deposit of $54 arrives. A second fee of $34 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "228", "answer_0": "246", "answer_3": "261", "answer_1": "216"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1509 per year. It increases by 15% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1710.35", "answer_0": "1721.35", "answer_1": "1735.35", "answer_2": "1282.65"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $594. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 09:15 to 12:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h15", "dur_0": "2h45", "dur_3": "2h15", "dur_1": "3h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $10k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 244:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_3": "16", "answer_0": "40", "answer_1": "26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $37k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed October 25, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 08", "date_3": "November 06", "date_1": "November 04", "date_0": "November 06"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_1": "37", "answer_3": "42", "answer_2": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $333. A fee of $131 is charged, then a deposit of $58 arrives. A second fee of $65 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "195", "answer_3": "178", "answer_1": "176", "answer_2": "192"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $986 per year. It increases by 20% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1138.2", "answer_2": "788.8", "answer_3": "1148.2", "answer_1": "1183.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 273 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "81", "answer_0": "72", "answer_1": "91", "answer_2": "53"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 761:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = $836; priority = low.\n\nClaim to verify: \"the amount is $836\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, September 02, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $722. A fee of $171 is charged, then a deposit of $73 arrives. A second fee of $85 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "541", "answer_0": "539", "answer_2": "563", "answer_1": "554"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = 2026-01-01; priority = low.\n\nClaim to verify: \"the date is 2026-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: status = open; priority = high.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = 2026-11-23; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 663:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $473. A fee of $180 is charged, then a deposit of $64 arrives. A second fee of $90 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "262", "answer_1": "272", "answer_0": "267", "answer_2": "279"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Contract signed July 08, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 14", "date_0": "July 16", "date_1": "July 14", "date_3": "July 16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 5 grams of salt per serving and serves 6 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1659 per year. It increases by 25% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2073.75", "answer_0": "2051.75", "answer_3": "2091.75", "answer_2": "1244.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.250", "p1": "0.500", "p0": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 12 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = 2026-04-17; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, October 12, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1990. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "37", "age_1": "34", "age_2": "36", "age_0": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1744 per year. It increases by 25% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2180", "answer_2": "1308", "answer_3": "2105", "answer_0": "2132"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 383 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "43", "answer_2": "29", "answer_0": "40", "answer_3": "48"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, May 23, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 377:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "56", "answer_0": "72", "answer_2": "16", "answer_1": "23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.417", "p3": "0.083", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 9 grams of salt per serving and serves 10 people. A shopper bought a 90 gram box and a 900 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $266. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Thursday, July 23, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "49", "answer_2": "14", "answer_3": "35", "answer_1": "19"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.333", "p2": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1015 per year. It increases by 15% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1118.25", "answer_2": "862.75", "answer_0": "1133.25", "answer_1": "1167.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 342:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 10 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "15", "answer_2": "10", "answer_0": "25", "answer_1": "13"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 322 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "236", "answer_2": "229", "answer_0": "216", "answer_3": "205"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $4k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1979. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "44", "age_2": "46", "age_0": "45", "age_3": "43"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = $716; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $815. A fee of $66 is charged, then a deposit of $49 arrives. A second fee of $33 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "788", "answer_1": "763", "answer_0": "765", "answer_2": "775"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 800:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = $860; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 327 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "106", "answer_3": "101", "answer_0": "120", "answer_2": "101"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=139, unit_price=156, discount_pct=0, region=EU, sku_total=295 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "21823", "v1": "295", "v0": "21684", "v2": "2168"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Contract signed April 22, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 02", "date_2": "May 02", "date_0": "May 04", "date_1": "April 30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=887, unit_price=677, discount_pct=0, region=EU, sku_total=1564 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1564", "v2": "60049", "v3": "599612", "v0": "600499"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $626. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=977, unit_price=855, discount_pct=0, region=EU, sku_total=1832 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1832", "v2": "83533", "v3": "836312", "v0": "835335"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $16k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = 2026-11-07; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = 2026-02-26; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=299, unit_price=627, discount_pct=0, region=EU, sku_total=926 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "187473", "v1": "926", "v3": "187772", "v2": "18747"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed October 19, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 28", "date_2": "November 01", "date_3": "October 30", "date_0": "October 30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 09:00 to 17:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "7h30", "dur_1": "8h00", "dur_0": "7h15", "dur_3": "7h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 50 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "66", "answer_1": "88", "answer_0": "72", "answer_3": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.375", "p1": "0.625", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 4 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, February 25, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.833", "p0": "0.667", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed September 07, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 19", "date_2": "September 23", "date_1": "September 17", "date_0": "September 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p3": "0.125", "p1": "0.500", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Thursday, May 28, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 360 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "183", "answer_2": "163", "answer_3": "190", "answer_0": "180"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=490, unit_price=549, discount_pct=0, region=EU, sku_total=1039 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "269010", "v1": "1039", "v3": "269500", "v2": "26901"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $508. A fee of $134 is charged, then a deposit of $66 arrives. A second fee of $67 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "346", "answer_1": "393", "answer_0": "373", "answer_3": "397"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p3": "0.444", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=414, unit_price=919, discount_pct=0, region=EU, sku_total=1333 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "38046", "v3": "380052", "v0": "380466", "v1": "1333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 114 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_1": "124", "answer_2": "102", "answer_3": "104"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 9 grams of salt per serving and serves 11 people. A shopper bought a 99 gram box and a 990 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed April 08, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 16", "date_1": "April 14", "date_3": "April 16", "date_2": "April 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p2": "0.500", "p3": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1880 per year. It increases by 20% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2246", "answer_3": "2240", "answer_1": "2256", "answer_2": "1504"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 3 grams of salt per serving and serves 11 people. A shopper bought a 33 gram box and a 330 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 135 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "109", "answer_0": "120", "answer_3": "129", "answer_1": "116"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p0": "0.583", "p2": "0.417", "p3": "0.545"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 9 grams of salt per serving and serves 7 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = $610; priority = high.\n\nClaim to verify: \"the status is $610\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1691 per year. It increases by 10% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1521.9", "answer_3": "1868.1", "answer_0": "1831.1", "answer_1": "1860.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1026 per year. It increases by 20% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1186.2", "answer_1": "1231.2", "answer_0": "1220.2", "answer_2": "820.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 442:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 130 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "82", "answer_0": "72", "answer_3": "64", "answer_1": "85"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = $85; priority = low.\n\nClaim to verify: \"the date is $85\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "17", "answer_0": "45", "answer_3": "35", "answer_2": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.545", "p0": "0.455", "p1": "0.545", "p3": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "20", "answer_0": "40", "answer_2": "15", "answer_3": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=438, unit_price=278, discount_pct=0, region=EU, sku_total=716 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "121764", "v3": "122202", "v1": "716", "v2": "12176"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p1": "0.667", "p0": "0.333", "p2": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = $986; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.286", "p2": "0.714", "p3": "0.167", "p1": "0.714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 395 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "75", "answer_2": "67", "answer_1": "78", "answer_0": "80"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "21", "answer_3": "9", "answer_2": "18", "answer_0": "27"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1959. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "68", "age_3": "69", "age_0": "67", "age_1": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, December 01, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "15", "answer_2": "12", "answer_0": "30", "answer_3": "18"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1979. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "43", "age_0": "45", "age_1": "44", "age_2": "46"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Wednesday, March 04, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $16k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 82 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "240", "answer_1": "243", "answer_2": "238", "answer_3": "253"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 2 grams of salt per serving and serves 11 people. A shopper bought a 22 gram box and a 220 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, February 23, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1989. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "35", "age_1": "34", "age_2": "36", "age_3": "37"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $397. A fee of $141 is charged, then a deposit of $25 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "196", "answer_1": "205", "answer_0": "211", "answer_3": "217"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed October 12, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 24", "date_2": "October 24", "date_0": "October 26", "date_1": "October 22"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $274 per year. It increases by 10% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "301.4", "answer_0": "268.4", "answer_3": "230.4", "answer_2": "246.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 06:15 to 13:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "6h00", "dur_0": "6h45", "dur_1": "7h00", "dur_2": "7h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $548. A fee of $126 is charged, then a deposit of $56 arrives. A second fee of $63 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "415", "answer_3": "426", "answer_1": "424", "answer_2": "407"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $546 per year. It increases by 5% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "556.3", "answer_3": "532.3", "answer_1": "573.3", "answer_2": "518.7"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = 2026-08-02; priority = low.\n\nClaim to verify: \"the amount is 2026-08-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 09:00 to 19:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "9h45", "dur_1": "10h15", "dur_0": "9h30", "dur_3": "9h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = 2026-05-13; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p3": "0.467", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = 2026-06-18; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1105 per year. It increases by 5% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1135.25", "answer_2": "1049.75", "answer_1": "1160.25", "answer_3": "1151.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 322:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "53", "age_0": "52", "age_3": "50", "age_1": "51"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1426 per year. It increases by 20% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1650.2", "answer_2": "1140.8", "answer_0": "1687.2", "answer_1": "1711.2"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.444", "p0": "0.556", "p1": "0.444", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.667", "p0": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 960:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1159 per year. It increases by 25% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1448.75", "answer_2": "869.25", "answer_0": "1404.75", "answer_3": "1409.75"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1965. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "61", "age_0": "60", "age_1": "59", "age_3": "58"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Saturday, October 17, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=113, unit_price=360, discount_pct=0, region=EU, sku_total=473 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "40567", "v1": "473", "v0": "40680", "v2": "4068"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 364 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "56", "answer_2": "79", "answer_3": "57", "answer_0": "72"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.286", "p0": "0.375", "p2": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 237:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.417", "p1": "0.538", "p0": "0.462", "p2": "0.538"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1388 per year. It increases by 15% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1179.8", "answer_1": "1596.2", "answer_3": "1558.2", "answer_0": "1548.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.714", "p1": "0.250", "p0": "0.750", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, October 05, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 253 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "82", "answer_1": "75", "answer_2": "92", "answer_0": "72"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $10k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $670. A fee of $177 is charged, then a deposit of $70 arrives. A second fee of $88 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "493", "answer_1": "473", "answer_0": "475", "answer_3": "504"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 368:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "28", "answer_0": "49", "answer_1": "31", "answer_3": "21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = 2026-05-02; priority = high.\n\nClaim to verify: \"the date is 2026-05-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $788. A fee of $41 is charged, then a deposit of $69 arrives. A second fee of $20 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "814", "answer_0": "796", "answer_2": "771", "answer_3": "789"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, August 07, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Sunday, September 20, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.429", "p3": "0.538", "p2": "0.429", "p0": "0.571"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_3": "32", "answer_0": "48", "answer_1": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=938, unit_price=698, discount_pct=0, region=EU, sku_total=1636 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1636", "v2": "65472", "v0": "654724", "v3": "653786"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $569. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $912 per year. It increases by 5% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "954.6", "answer_0": "929.6", "answer_1": "957.6", "answer_2": "866.4"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p0": "0.500", "p2": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=833, unit_price=440, discount_pct=0, region=EU, sku_total=1273 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "36652", "v1": "1273", "v0": "366520", "v3": "367353"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed December 19, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 21", "date_1": "December 22", "date_0": "December 23", "date_3": "December 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, January 25, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = $854; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 10:15 to 20:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "10h15", "dur_3": "9h15", "dur_2": "10h00", "dur_0": "9h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=293, unit_price=899, discount_pct=0, region=EU, sku_total=1192 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "263407", "v1": "1192", "v3": "263700", "v2": "26340"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.083", "p2": "0.250", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1217 per year. It increases by 25% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1521.25", "answer_3": "1469.25", "answer_2": "912.75", "answer_0": "1482.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $508. A fee of $160 is charged, then a deposit of $73 arrives. A second fee of $80 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "341", "answer_1": "328", "answer_2": "315", "answer_3": "318"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.444", "p3": "0.500", "p0": "0.556", "p2": "0.444"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 246 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_1": "70", "answer_3": "59", "answer_2": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=597, unit_price=905, discount_pct=0, region=EU, sku_total=1502 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "539688", "v0": "540285", "v2": "54028", "v1": "1502"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=941, unit_price=285, discount_pct=0, region=EU, sku_total=1226 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "26818", "v0": "268185", "v3": "269126", "v1": "1226"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 8 grams of salt per serving and serves 12 people. A shopper bought a 96 gram box and a 960 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = 2026-12-24; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.667", "p2": "0.286", "p1": "0.286", "p0": "0.714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_1": "22", "answer_0": "50", "answer_2": "15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "31", "answer_3": "30", "answer_0": "55", "answer_2": "25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = ops; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $937 per year. It increases by 20% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1107.4", "answer_1": "1124.4", "answer_2": "749.6", "answer_3": "1113.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1975. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "52", "age_2": "51", "age_0": "50", "age_1": "49"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 375 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_2": "76", "answer_3": "75", "answer_1": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.417", "p1": "0.583", "p2": "0.583", "p3": "0.364"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $636. A fee of $104 is charged, then a deposit of $63 arrives. A second fee of $52 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "555", "answer_1": "530", "answer_0": "543", "answer_2": "546"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = $287; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $2000 per year. It increases by 20% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2390", "answer_3": "2410", "answer_1": "2400", "answer_2": "1600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_1": "25", "answer_3": "21", "answer_0": "39"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = $864; priority = low.\n\nClaim to verify: \"the status is $864\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Wednesday, February 04, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed April 01, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 08", "date_1": "April 06", "date_2": "April 10", "date_3": "April 08"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 245:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, November 05, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 07:00 to 08:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "1h00", "dur_2": "1h15", "dur_3": "0h15", "dur_1": "1h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1988. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "34", "age_0": "36", "age_1": "35", "age_2": "37"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $457. A fee of $142 is charged, then a deposit of $48 arrives. A second fee of $71 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "268", "answer_2": "305", "answer_0": "292", "answer_1": "275"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $9k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = 2026-03-24; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 851:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p1": "0.250", "p3": "0.050", "p2": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_1": "23", "answer_0": "44", "answer_3": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = 2026-05-08; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed November 06, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 13", "date_0": "November 13", "date_1": "November 11", "date_2": "November 11"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.556", "p2": "0.444", "p1": "0.444", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = 2026-12-01; priority = low.\n\nClaim to verify: \"the amount is 2026-12-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.125", "p1": "0.375", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.167", "p2": "0.167", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = pending; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p1": "0.600", "p2": "0.600", "p3": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = 2026-12-09; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.125", "p1": "0.375", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "21", "answer_0": "33", "answer_2": "15", "answer_3": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed June 13, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 16", "date_2": "June 19", "date_3": "June 18", "date_0": "June 17"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.444", "p3": "0.500", "p2": "0.444", "p0": "0.556"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $36k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p3": "0.050", "p1": "0.200", "p0": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.571", "p1": "0.375", "p2": "0.375", "p0": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=911, unit_price=710, discount_pct=0, region=EU, sku_total=1621 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "64681", "v1": "1621", "v0": "646810", "v3": "647721"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 8 grams of salt per serving and serves 12 people. A shopper bought a 96 gram box and a 960 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 377:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A shift runs from 19:30 to 20:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h30", "dur_2": "0h45", "dur_1": "0h30", "dur_0": "0h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p0": "0.150", "p3": "0.050", "p2": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = 2026-01-18; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.533", "p3": "0.429", "p1": "0.533", "p0": "0.467"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.150", "p0": "0.200", "p1": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.545", "p2": "0.455", "p3": "0.500", "p1": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = 2026-02-26; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p0": "0.200", "p3": "0.050", "p1": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed February 27, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 09", "date_2": "March 13", "date_1": "March 07", "date_0": "March 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed October 14, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 21", "date_2": "October 25", "date_3": "October 23", "date_0": "October 23"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "13", "answer_0": "21", "answer_3": "12", "answer_2": "9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $748. A fee of $83 is charged, then a deposit of $42 arrives. A second fee of $41 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "671", "answer_0": "666", "answer_1": "636", "answer_3": "674"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 12 people. A shopper bought a 84 gram box and a 840 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $475. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1215 per year. It increases by 5% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1251.75", "answer_2": "1154.25", "answer_3": "1261.75", "answer_1": "1275.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $587 per year. It increases by 10% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "528.3", "answer_0": "607.7", "answer_1": "645.7", "answer_3": "639.7"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, March 18, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 9 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.625", "p1": "0.625", "p3": "0.286", "p0": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = $845; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 5 grams of salt per serving and serves 9 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1761 per year. It increases by 20% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2113.2", "answer_0": "2068.2", "answer_3": "2030.2", "answer_2": "1408.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = 2026-04-11; priority = low.\n\nClaim to verify: \"the status is 2026-04-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, March 15, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $231. A fee of $159 is charged, then a deposit of $30 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "40", "answer_1": "7", "answer_2": "37", "answer_0": "23"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "10", "answer_3": "35", "answer_0": "45", "answer_1": "17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 15:30 to 22:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "7h00", "dur_0": "6h30", "dur_1": "6h45", "dur_3": "5h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = 2026-03-02; priority = low.\n\nClaim to verify: \"the owner is 2026-03-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $836. A fee of $169 is charged, then a deposit of $63 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "646", "answer_3": "666", "answer_2": "629", "answer_1": "631"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=539, unit_price=838, discount_pct=0, region=EU, sku_total=1377 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "451143", "v2": "45168", "v1": "1377", "v0": "451682"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=912, unit_price=553, discount_pct=0, region=EU, sku_total=1465 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1465", "v0": "504336", "v2": "50433", "v3": "503424"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = 2026-09-22; priority = low.\n\nClaim to verify: \"the amount is 2026-09-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1964. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "61", "age_0": "62", "age_2": "63", "age_3": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = open; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1960. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "68", "age_1": "65", "age_2": "67", "age_0": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1989. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "34", "age_3": "37", "age_0": "35", "age_2": "36"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Wednesday, July 01, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p2": "0.800", "p3": "0.111", "p1": "0.800"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = ops; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $810. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = 2026-12-23; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $33k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = 2026-02-03; priority = low.\n\nClaim to verify: \"the owner is 2026-02-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 281:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, June 16, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $289. A fee of $55 is charged, then a deposit of $38 arrives. A second fee of $27 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "260", "answer_2": "218", "answer_3": "263", "answer_0": "245"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1763 per year. It increases by 10% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1926.3", "answer_1": "1939.3", "answer_2": "1586.7", "answer_3": "1899.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 993:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.727", "p2": "0.727", "p3": "0.200", "p0": "0.273"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = $645; priority = low.\n\nClaim to verify: \"the amount is $645\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed January 21, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 01", "date_0": "February 03", "date_1": "January 30", "date_3": "February 01"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=786, unit_price=125, discount_pct=0, region=EU, sku_total=911 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "911", "v2": "9825", "v0": "98250", "v3": "99036"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 129 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "52", "answer_3": "22", "answer_1": "44", "answer_0": "40"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 352 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "105", "answer_3": "92", "answer_2": "93", "answer_0": "108"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.600", "p2": "0.333", "p0": "0.667", "p1": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.250", "p0": "0.200", "p2": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A recipe needs 3 grams of salt per serving and serves 9 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = closed; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $819 per year. It increases by 5% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "815.95", "answer_3": "827.95", "answer_2": "778.05", "answer_1": "859.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed June 02, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 14", "date_2": "June 14", "date_0": "June 16", "date_1": "June 12"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.545", "p3": "0.400", "p2": "0.545", "p0": "0.455"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 18, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 28", "date_3": "April 30", "date_2": "April 29", "date_0": "May 01"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=190, unit_price=609, discount_pct=0, region=EU, sku_total=799 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "799", "v3": "115520", "v0": "115710", "v2": "11571"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $31k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 182 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "64", "answer_1": "77", "answer_3": "68", "answer_2": "44"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $241. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 10:00 to 22:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "12h00", "dur_0": "11h45", "dur_1": "12h30", "dur_3": "11h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $39k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 12 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 07:30 to 21:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "13h45", "dur_3": "12h45", "dur_0": "13h30", "dur_2": "13h45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 10 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.556", "p2": "0.556", "p3": "0.375", "p0": "0.444"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_2": "16", "answer_0": "28", "answer_1": "19"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1155 per year. It increases by 10% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1270.5", "answer_3": "1209.5", "answer_2": "1039.5", "answer_0": "1228.5"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 172 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "89", "answer_0": "96", "answer_3": "94", "answer_2": "77"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed December 26, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 03", "date_0": "January 05", "date_1": "January 02", "date_3": "January 04"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = 2026-03-24; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = $226; priority = high.\n\nClaim to verify: \"the owner is $226\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed June 07, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 16", "date_0": "June 18", "date_3": "June 18", "date_2": "June 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=678, unit_price=416, discount_pct=0, region=EU, sku_total=1094 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "28204", "v0": "282048", "v1": "1094", "v3": "281370"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.500", "p0": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1966. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "61", "age_0": "60", "age_3": "62", "age_1": "59"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = $780; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $494. A fee of $162 is charged, then a deposit of $59 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "283", "answer_1": "325", "answer_3": "336", "answer_0": "310"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $897. A fee of $39 is charged, then a deposit of $23 arrives. A second fee of $19 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "838", "answer_1": "879", "answer_0": "862", "answer_2": "842"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 83 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "108", "answer_2": "81", "answer_1": "82", "answer_0": "98"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=721, unit_price=341, discount_pct=0, region=EU, sku_total=1062 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "245140", "v2": "24586", "v1": "1062", "v0": "245861"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = 2026-08-06; priority = high.\n\nClaim to verify: \"the date is 2026-08-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $935 per year. It increases by 25% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "701.25", "answer_0": "1142.75", "answer_1": "1168.75", "answer_3": "1152.75"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 119 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "308", "answer_1": "298", "answer_0": "294", "answer_3": "314"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "10", "answer_0": "35", "answer_1": "27", "answer_2": "25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=650, unit_price=988, discount_pct=0, region=EU, sku_total=1638 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "642200", "v3": "641550", "v1": "1638", "v2": "64220"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1978. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "46", "age_0": "47", "age_2": "48", "age_3": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1722 per year. It increases by 5% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1774.1", "answer_3": "1784.1", "answer_1": "1808.1", "answer_2": "1635.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed December 05, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 17", "date_2": "December 16", "date_1": "December 15", "date_0": "December 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.500", "p2": "0.375", "p1": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1084 per year. It increases by 20% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1324.8", "answer_0": "1286.8", "answer_1": "1300.8", "answer_2": "867.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "35", "answer_3": "21", "answer_1": "17", "answer_2": "14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = 2026-06-10; priority = high.\n\nClaim to verify: \"the date is 2026-06-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.333", "p1": "0.500", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $861. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = 2026-01-24; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.600", "p3": "0.333", "p1": "0.600", "p0": "0.400"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $31k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=594, unit_price=117, discount_pct=0, region=EU, sku_total=711 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "69498", "v1": "711", "v3": "68904", "v2": "6949"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, May 17, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = $201; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 96 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "93", "answer_3": "98", "answer_0": "105", "answer_2": "108"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_0": "42", "answer_3": "24", "answer_1": "22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = 2026-03-10; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 642:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 17:30 to 20:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "2h15", "dur_0": "2h00", "dur_1": "2h45", "dur_3": "1h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 16:45 to 19:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h45", "dur_3": "1h45", "dur_0": "2h00", "dur_2": "2h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed March 11, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 16", "date_0": "March 18", "date_2": "March 16", "date_3": "March 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.700", "p1": "0.300", "p2": "0.300", "p3": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $753 per year. It increases by 15% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "865.95", "answer_0": "855.95", "answer_2": "640.05", "answer_3": "863.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.333", "p3": "0.625", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 76 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "46", "answer_3": "17", "answer_2": "35", "answer_0": "32"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.750", "p2": "0.222", "p1": "0.222", "p0": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.417", "p0": "0.333", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=539, unit_price=416, discount_pct=0, region=EU, sku_total=955 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "223685", "v0": "224224", "v2": "22422", "v1": "955"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.583", "p3": "0.364", "p0": "0.417", "p1": "0.583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=913, unit_price=607, discount_pct=0, region=EU, sku_total=1520 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1520", "v0": "554191", "v3": "553278", "v2": "55419"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1979. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "47", "age_3": "49", "age_1": "46", "age_2": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "20", "answer_1": "27", "answer_3": "28", "answer_0": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $213 per year. It increases by 20% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "255.6", "answer_3": "216.6", "answer_0": "205.6", "answer_2": "170.4"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $876. A fee of $63 is charged, then a deposit of $77 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "849", "answer_3": "861", "answer_0": "859", "answer_1": "840"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = 2026-04-02; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.500", "p3": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 748:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_3": "6", "answer_0": "24", "answer_1": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=987, unit_price=689, discount_pct=0, region=EU, sku_total=1676 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1676", "v0": "680043", "v2": "68004", "v3": "681030"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "16", "answer_0": "32", "answer_2": "16", "answer_1": "20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.400", "p3": "0.556", "p1": "0.400", "p0": "0.600"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1961. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "62", "age_2": "64", "age_0": "63", "age_3": "65"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.500", "p2": "0.167", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $424 per year. It increases by 20% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "339.2", "answer_0": "467.8", "answer_3": "499.8", "answer_1": "508.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $439. A fee of $54 is charged, then a deposit of $53 arrives. A second fee of $27 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "411", "answer_1": "403", "answer_3": "395", "answer_2": "403"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p3": "0.333", "p2": "0.571", "p0": "0.429"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $426. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "20", "answer_3": "16", "answer_1": "24", "answer_0": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=397, unit_price=909, discount_pct=0, region=EU, sku_total=1306 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "360873", "v3": "361270", "v1": "1306", "v2": "36087"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=167, unit_price=310, discount_pct=0, region=EU, sku_total=477 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "51770", "v3": "51603", "v2": "5177", "v1": "477"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = $554; priority = high.\n\nClaim to verify: \"the amount is $554\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = 2026-12-14; priority = high.\n\nClaim to verify: \"the status is 2026-12-14\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = 2026-12-07; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 08:00 to 09:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "1h45", "dur_0": "1h30", "dur_3": "0h45", "dur_2": "2h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.273", "p3": "0.700", "p0": "0.727", "p1": "0.273"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1969. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "54", "age_0": "55", "age_3": "53", "age_2": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 6 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = 2026-12-13; priority = high.\n\nClaim to verify: \"the amount is 2026-12-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.429", "p0": "0.571", "p3": "0.538", "p1": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1972. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "54", "age_0": "52", "age_2": "53", "age_1": "51"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 53 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "16", "answer_3": "56", "answer_2": "44", "answer_0": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 2 grams of salt per serving and serves 9 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 935:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p2": "0.125", "p0": "0.250", "p3": "0.125"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=274, unit_price=999, discount_pct=0, region=EU, sku_total=1273 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1273", "v2": "27372", "v0": "273726", "v3": "274000"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1970. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "55", "age_1": "54", "age_3": "53", "age_2": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 532:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $255. A fee of $198 is charged, then a deposit of $46 arrives. A second fee of $99 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "34", "answer_0": "4", "answer_3": "-22", "answer_2": "-16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = $758; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Saturday, December 26, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $605 per year. It increases by 10% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "684.5", "answer_2": "544.5", "answer_0": "654.5", "answer_1": "665.5"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.636", "p0": "0.364", "p2": "0.636", "p3": "0.300"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.462", "p3": "0.417", "p1": "0.538", "p2": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = $508; priority = high.\n\nClaim to verify: \"the amount is $508\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed July 22, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 27", "date_2": "July 29", "date_1": "July 25", "date_0": "July 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $546. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, March 02, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = $841; priority = low.\n\nClaim to verify: \"the status is $841\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $846; priority = low.\n\nClaim to verify: \"the owner is $846\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 301:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.583", "p3": "0.364", "p0": "0.417", "p2": "0.583"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 671:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p0": "0.400", "p2": "0.600", "p1": "0.600"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 325 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_2": "58", "answer_3": "72", "answer_1": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_3": "20", "answer_2": "20", "answer_1": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 8 grams of salt per serving and serves 7 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=413, unit_price=306, discount_pct=0, region=EU, sku_total=719 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "126378", "v1": "719", "v3": "125965", "v2": "12637"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $320 per year. It increases by 25% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "371", "answer_2": "240", "answer_1": "400", "answer_3": "409"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed June 15, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 19", "date_2": "June 21", "date_0": "June 19", "date_3": "June 21"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=480, unit_price=733, discount_pct=0, region=EU, sku_total=1213 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "351840", "v1": "1213", "v2": "35184", "v3": "352320"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=774, unit_price=508, discount_pct=0, region=EU, sku_total=1282 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1282", "v0": "393192", "v2": "39319", "v3": "393966"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.333", "p3": "0.083", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.714", "p3": "0.667", "p1": "0.286", "p2": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "32", "answer_0": "72", "answer_3": "40", "answer_1": "37"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 899:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "14", "answer_3": "15", "answer_2": "9", "answer_0": "24"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "16", "answer_0": "32", "answer_2": "16", "answer_1": "20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.455", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 847:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.167", "p1": "0.333", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = 2026-04-18; priority = low.\n\nClaim to verify: \"the status is 2026-04-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1694 per year. It increases by 15% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1439.9", "answer_1": "1948.1", "answer_0": "1916.1", "answer_3": "1922.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = open; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = 2026-07-12; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: owner = $21; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 7 grams of salt per serving and serves 8 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 5 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1963. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "63", "age_1": "62", "age_2": "64", "age_3": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_0": "48", "answer_1": "20", "answer_3": "32"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1966. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_2": "59", "age_3": "56", "age_1": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 201:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed May 17, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 28", "date_3": "May 28", "date_2": "May 30", "date_1": "May 26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.125", "p0": "0.375", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = closed; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.417", "p2": "0.583", "p1": "0.583", "p3": "0.364"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $403 per year. It increases by 10% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "445.3", "answer_2": "362.7", "answer_1": "443.3", "answer_0": "412.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, December 14, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = $798; priority = high.\n\nClaim to verify: \"the owner is $798\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = 2026-05-12; priority = low.\n\nClaim to verify: \"the amount is 2026-05-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $675. A fee of $170 is charged, then a deposit of $86 arrives. A second fee of $85 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "487", "answer_3": "524", "answer_0": "506", "answer_1": "516"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p2": "0.500", "p3": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $343. A fee of $122 is charged, then a deposit of $45 arrives. A second fee of $61 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "205", "answer_1": "217", "answer_3": "201", "answer_2": "215"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=178, unit_price=147, discount_pct=0, region=EU, sku_total=325 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "26166", "v3": "25988", "v1": "325", "v2": "2616"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 206 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "64", "answer_3": "58", "answer_1": "54", "answer_2": "82"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p2": "0.500", "p3": "0.167", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $691. A fee of $44 is charged, then a deposit of $21 arrives. A second fee of $22 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "670", "answer_0": "646", "answer_1": "625", "answer_3": "633"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed September 11, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 25", "date_1": "September 19", "date_0": "September 23", "date_3": "September 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, January 18, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "18", "answer_3": "8", "answer_0": "24", "answer_2": "16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.286", "p0": "0.714", "p1": "0.286", "p3": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "63", "answer_3": "21", "answer_2": "42", "answer_1": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 977:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $830. A fee of $154 is charged, then a deposit of $85 arrives. A second fee of $77 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "672", "answer_1": "673", "answer_3": "687", "answer_0": "684"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed December 06, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 17", "date_2": "December 19", "date_3": "December 17", "date_1": "December 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = ops; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Sunday, January 18, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Thursday, January 08, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 4 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $792 per year. It increases by 15% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "851.8", "answer_2": "673.2", "answer_0": "876.8", "answer_1": "910.8"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p0": "0.100", "p1": "0.150", "p3": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $13k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.333", "p1": "0.500", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 5 grams of salt per serving and serves 10 people. A shopper bought a 50 gram box and a 500 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p1": "0.600", "p0": "0.400", "p2": "0.600"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = $167; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1968. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "57", "age_3": "55", "age_2": "58", "age_1": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 06:45 to 22:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "15h30", "dur_3": "14h15", "dur_0": "15h00", "dur_1": "15h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1974. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "53", "age_1": "51", "age_3": "54", "age_0": "52"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 10 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 227:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, December 16, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed December 23, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 29", "date_1": "December 27", "date_3": "December 29", "date_2": "December 31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.083", "p0": "0.250", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=637, unit_price=595, discount_pct=0, region=EU, sku_total=1232 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "379015", "v1": "1232", "v2": "37901", "v3": "378378"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $969 per year. It increases by 20% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1118.8", "answer_1": "1162.8", "answer_0": "1126.8", "answer_2": "775.2"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.167", "p3": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 15:45 to 22:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "5h30", "dur_2": "5h45", "dur_3": "5h15", "dur_1": "6h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = 2026-12-11; priority = low.\n\nClaim to verify: \"the owner is 2026-12-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1966. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "59", "age_3": "58", "age_0": "60", "age_2": "61"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = closed; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=524, unit_price=234, discount_pct=0, region=EU, sku_total=758 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "123140", "v0": "122616", "v1": "758", "v2": "12261"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "63", "answer_2": "35", "answer_1": "39", "answer_3": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "56", "answer_2": "48", "answer_0": "104", "answer_1": "55"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_1": "25", "answer_3": "42", "answer_0": "60"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1613 per year. It increases by 20% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1907.6", "answer_0": "1902.6", "answer_2": "1290.4", "answer_1": "1935.6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=147, unit_price=814, discount_pct=0, region=EU, sku_total=961 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "961", "v0": "119658", "v3": "119511", "v2": "11965"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.250", "p1": "0.333", "p3": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p1": "0.500", "p3": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = $401; priority = high.\n\nClaim to verify: \"the amount is $401\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.833", "p3": "0.167", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $37k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = open; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 08:15 to 16:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "8h30", "dur_2": "8h15", "dur_3": "7h30", "dur_0": "7h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = 2026-07-24; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 303 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "168", "answer_1": "161", "answer_3": "171", "answer_2": "151"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=915, unit_price=261, discount_pct=0, region=EU, sku_total=1176 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1176", "v0": "238815", "v3": "237900", "v2": "23881"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, November 02, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Wednesday, March 04, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1152 per year. It increases by 5% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1197.6", "answer_2": "1094.4", "answer_3": "1217.6", "answer_1": "1209.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 206 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "195", "answer_2": "166", "answer_1": "166", "answer_0": "180"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, January 07, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p3": "0.050", "p0": "0.150", "p2": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.500", "p0": "0.375", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $285. A fee of $109 is charged, then a deposit of $33 arrives. A second fee of $54 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "155", "answer_3": "177", "answer_1": "148", "answer_2": "170"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = closed; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1961. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "65", "age_1": "62", "age_0": "63", "age_2": "64"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed August 04, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 16", "date_0": "August 18", "date_1": "August 14", "date_2": "August 20"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.250", "p3": "0.083", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p0": "0.167", "p3": "0.083", "p1": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1974. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "52", "age_1": "50", "age_0": "51", "age_3": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 132 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "200", "answer_1": "187", "answer_3": "208", "answer_2": "206"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 292 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "152", "answer_0": "144", "answer_3": "146", "answer_2": "164"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1986. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "37", "age_2": "40", "age_0": "39", "age_1": "38"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = 2026-08-18; priority = low.\n\nClaim to verify: \"the status is 2026-08-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=866, unit_price=543, discount_pct=0, region=EU, sku_total=1409 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "470238", "v2": "47023", "v1": "1409", "v3": "469372"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.500", "p1": "0.625", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: date = 2026-07-03; priority = low.\n\nClaim to verify: \"the date is 2026-07-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = pending; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p2": "0.250", "p0": "0.333", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 54 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "35", "answer_3": "29", "answer_1": "60", "answer_0": "48"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 10:15 to 13:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "3h00", "dur_1": "3h15", "dur_3": "2h15", "dur_2": "3h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = open; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $721. A fee of $168 is charged, then a deposit of $82 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "573", "answer_1": "549", "answer_0": "551", "answer_2": "561"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "17", "answer_0": "25", "answer_2": "15", "answer_3": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = $634; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 341 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_2": "86", "answer_1": "64", "answer_3": "76"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.250", "p2": "0.125", "p1": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.167", "p0": "0.250", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $420. A fee of $50 is charged, then a deposit of $71 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "392", "answer_0": "416", "answer_3": "408", "answer_1": "445"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $216. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "9", "answer_0": "21", "answer_2": "12", "answer_1": "15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p1": "0.500", "p3": "0.455"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $31k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Friday, July 17, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 2 grams of salt per serving and serves 6 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1839 per year. It increases by 5% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1889.95", "answer_3": "1875.95", "answer_2": "1747.05", "answer_1": "1930.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1036 per year. It increases by 10% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "932.4", "answer_0": "1125.6", "answer_3": "1092.6", "answer_1": "1139.6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = open; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 07:15 to 15:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "8h15", "dur_2": "8h15", "dur_3": "7h15", "dur_0": "8h00"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 9 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=541, unit_price=704, discount_pct=0, region=EU, sku_total=1245 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "38086", "v1": "1245", "v0": "380864", "v3": "380323"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = 2026-03-02; priority = high.\n\nClaim to verify: \"the date is 2026-03-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.083", "p2": "0.250", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 286:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1828 per year. It increases by 15% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2057.2", "answer_1": "2102.2", "answer_0": "2070.2", "answer_2": "1553.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, September 26, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $772. A fee of $139 is charged, then a deposit of $76 arrives. A second fee of $69 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "631", "answer_0": "640", "answer_3": "649", "answer_2": "642"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 06:15 to 10:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "2h45", "dur_1": "3h45", "dur_0": "3h00", "dur_2": "3h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $840 per year. It increases by 5% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "877", "answer_1": "882", "answer_2": "798", "answer_0": "845"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1962. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "64", "age_1": "61", "age_0": "62", "age_2": "63"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1738 per year. It increases by 15% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1998.7", "answer_3": "1938.7", "answer_0": "1953.7", "answer_2": "1477.3"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Monday, April 13, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = 2026-08-24; priority = low.\n\nClaim to verify: \"the owner is 2026-08-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 10:45 to 12:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h15", "dur_3": "0h30", "dur_0": "0h45", "dur_1": "1h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1677 per year. It increases by 10% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1814.7", "answer_3": "1837.7", "answer_2": "1509.3", "answer_1": "1844.7"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p0": "0.667", "p3": "0.167", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1959. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "69", "age_2": "68", "age_1": "66", "age_0": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1958. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "66", "age_3": "64", "age_1": "65", "age_2": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, July 14, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 76 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "54", "answer_2": "52", "answer_3": "59", "answer_1": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 357 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_2": "66", "answer_1": "59", "answer_3": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1910 per year. It increases by 5% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1975.5", "answer_2": "1814.5", "answer_3": "2011.5", "answer_1": "2005.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p1": "0.500", "p3": "0.467"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 10:30 to 20:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "9h45", "dur_3": "8h45", "dur_2": "9h15", "dur_0": "9h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "11", "answer_3": "12", "answer_0": "20", "answer_2": "8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 656:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "15", "answer_3": "18", "answer_2": "12", "answer_0": "30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p2": "0.150", "p1": "0.250", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = $304; priority = high.\n\nClaim to verify: \"the amount is $304\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = closed; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1962. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "64", "age_2": "65", "age_3": "66", "age_1": "63"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = 2026-07-19; priority = high.\n\nClaim to verify: \"the status is 2026-07-19\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.364", "p1": "0.364", "p0": "0.636", "p3": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 12:00 to 21:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "9h00", "dur_3": "8h00", "dur_2": "9h15", "dur_0": "8h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $830. A fee of $156 is charged, then a deposit of $88 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "674", "answer_2": "672", "answer_3": "673", "answer_0": "684"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 100 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "108", "answer_3": "113", "answer_2": "105", "answer_0": "96"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $834. A fee of $170 is charged, then a deposit of $66 arrives. A second fee of $85 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "645", "answer_1": "637", "answer_3": "654", "answer_2": "639"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 589:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed October 16, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 28", "date_1": "October 26", "date_0": "October 30", "date_2": "November 01"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = 2026-02-23; priority = high.\n\nClaim to verify: \"the owner is 2026-02-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.750", "p0": "0.250", "p3": "0.143", "p1": "0.750"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $253. A fee of $56 is charged, then a deposit of $71 arrives. A second fee of $28 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "240", "answer_3": "262", "answer_2": "254", "answer_1": "237"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1985. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "42", "age_3": "39", "age_1": "40", "age_0": "41"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 12:15 to 21:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "8h30", "dur_2": "9h00", "dur_1": "8h45", "dur_3": "7h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 81 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_3": "62", "answer_2": "92", "answer_1": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 565:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $295 per year. It increases by 5% this year, then a $40 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "269.75", "answer_1": "309.75", "answer_3": "245.75", "answer_2": "280.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=671, unit_price=948, discount_pct=0, region=EU, sku_total=1619 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "636108", "v3": "635437", "v1": "1619", "v2": "63610"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "21", "answer_3": "9", "answer_1": "15", "answer_2": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 349 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_1": "52", "answer_2": "67", "answer_3": "44"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 578:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $861. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.400", "p0": "0.600", "p3": "0.556", "p1": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p1": "0.200", "p3": "0.050", "p2": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1022 per year. It increases by 5% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1062.1", "answer_2": "970.9", "answer_1": "1073.1", "answer_0": "1043.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 392 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "56", "answer_0": "54", "answer_2": "43", "answer_3": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "20", "answer_2": "15", "answer_0": "35", "answer_1": "19"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.429", "p2": "0.533", "p0": "0.467", "p1": "0.533"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p1": "0.200", "p0": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.667", "p3": "0.167", "p1": "0.833"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1975. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "49", "age_3": "47", "age_1": "48", "age_2": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Saturday, April 11, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=472, unit_price=311, discount_pct=0, region=EU, sku_total=783 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "14679", "v1": "783", "v3": "147264", "v0": "146792"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = 2026-11-08; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.083", "p1": "0.417", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $396 per year. It increases by 10% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "387.6", "answer_3": "372.6", "answer_1": "435.6", "answer_2": "356.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, July 07, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 158 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "157", "answer_3": "143", "answer_2": "155", "answer_0": "140"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=140, unit_price=993, discount_pct=0, region=EU, sku_total=1133 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "13902", "v3": "138880", "v1": "1133", "v0": "139020"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.167", "p2": "0.083", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=124, unit_price=676, discount_pct=0, region=EU, sku_total=800 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "8382", "v3": "83948", "v1": "800", "v0": "83824"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $115. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed August 12, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "August 17", "date_2": "August 15", "date_3": "August 17", "date_1": "August 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1963. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "64", "age_3": "61", "age_0": "63", "age_1": "62"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $128. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed May 01, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 13", "date_3": "May 11", "date_1": "May 09", "date_2": "May 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Friday, July 17, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=593, unit_price=102, discount_pct=0, region=EU, sku_total=695 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "59893", "v1": "695", "v2": "6048", "v0": "60486"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=757, unit_price=958, discount_pct=0, region=EU, sku_total=1715 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "724449", "v2": "72520", "v0": "725206", "v1": "1715"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed November 05, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 19", "date_2": "November 21", "date_1": "November 15", "date_3": "November 17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = open; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 279:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.273", "p2": "0.667", "p1": "0.667", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p3": "0.400", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 721:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = closed; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1441 per year. It increases by 15% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1657.15", "answer_0": "1644.15", "answer_2": "1224.85", "answer_3": "1654.15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, August 10, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = 2026-09-16; priority = high.\n\nClaim to verify: \"the date is 2026-09-16\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, June 17, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = 2026-10-14; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p3": "0.050", "p1": "0.200", "p0": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed October 20, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 27", "date_2": "October 27", "date_3": "October 29", "date_0": "October 29"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = 2026-03-18; priority = high.\n\nClaim to verify: \"the amount is 2026-03-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $631. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 219 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "72", "answer_2": "80", "answer_0": "90", "answer_1": "85"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 690:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $700. A fee of $49 is charged, then a deposit of $43 arrives. A second fee of $24 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "648", "answer_1": "698", "answer_0": "670", "answer_2": "673"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_3": "24", "answer_1": "51", "answer_2": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 973:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1986. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "37", "age_0": "38", "age_3": "40", "age_2": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 213 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "156", "answer_2": "176", "answer_0": "160", "answer_3": "163"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1988. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "37", "age_2": "38", "age_3": "39", "age_1": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "30", "answer_3": "14", "answer_0": "42", "answer_2": "28"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $531. A fee of $132 is charged, then a deposit of $78 arrives. A second fee of $66 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "385", "answer_0": "411", "answer_2": "419", "answer_1": "425"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, June 20, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = 2026-08-20; priority = low.\n\nClaim to verify: \"the amount is 2026-08-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.083", "p1": "0.333", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = $815; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.385", "p1": "0.385", "p3": "0.583", "p0": "0.615"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 2 grams of salt per serving and serves 8 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, June 13, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Thursday, January 15, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 938:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, July 22, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, August 28, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $316. A fee of $52 is charged, then a deposit of $21 arrives. A second fee of $26 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "279", "answer_1": "243", "answer_2": "268", "answer_0": "259"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = 2026-11-22; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $555. A fee of $181 is charged, then a deposit of $44 arrives. A second fee of $90 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "346", "answer_0": "328", "answer_1": "315", "answer_2": "309"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.500", "p3": "0.467", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "15", "answer_2": "10", "answer_0": "25", "answer_1": "13"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 84 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "168", "answer_1": "185", "answer_2": "148", "answer_3": "179"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 155 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "65", "answer_3": "92", "answer_2": "89", "answer_0": "80"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 08:00 to 09:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h45", "dur_0": "1h15", "dur_3": "0h45", "dur_1": "1h45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $361. A fee of $41 is charged, then a deposit of $65 arrives. A second fee of $20 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "369", "answer_1": "338", "answer_3": "360", "answer_0": "365"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 16:45 to 18:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h00", "dur_0": "1h45", "dur_2": "2h15", "dur_3": "1h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=926, unit_price=977, discount_pct=0, region=EU, sku_total=1903 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "903776", "v2": "90470", "v1": "1903", "v0": "904702"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $293. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1192 per year. It increases by 15% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1013.2", "answer_0": "1336.8", "answer_3": "1350.8", "answer_1": "1370.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.125", "p1": "0.625", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_0": "56", "answer_1": "44", "answer_2": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = $97; priority = low.\n\nClaim to verify: \"the status is $97\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = $304; priority = low.\n\nClaim to verify: \"the amount is $304\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = $752; priority = high.\n\nClaim to verify: \"the owner is $752\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.300", "p0": "0.700", "p2": "0.300", "p3": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1030 per year. It increases by 10% this year, then a $40 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1133", "answer_3": "1106", "answer_2": "927", "answer_0": "1093"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $747. A fee of $134 is charged, then a deposit of $66 arrives. A second fee of $67 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "593", "answer_3": "583", "answer_1": "629", "answer_0": "612"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $811. A fee of $133 is charged, then a deposit of $44 arrives. A second fee of $66 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "649", "answer_1": "636", "answer_0": "656", "answer_2": "668"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 12:15 to 17:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "5h00", "dur_3": "4h00", "dur_2": "5h00", "dur_0": "4h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = 2026-07-09; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1959. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "66", "age_1": "65", "age_3": "64", "age_2": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.800", "p0": "0.200", "p3": "0.111", "p2": "0.800"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.125", "p0": "0.375", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1965 per year. It increases by 5% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1866.75", "answer_3": "2020.25", "answer_0": "2038.25", "answer_1": "2063.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $746. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 888:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1334 per year. It increases by 5% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1400.7", "answer_0": "1355.7", "answer_3": "1375.7", "answer_2": "1267.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = $598; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, July 04, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = 2026-11-19; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 353:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed June 11, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 18", "date_2": "June 16", "date_0": "June 18", "date_1": "June 16"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $455. A fee of $192 is charged, then a deposit of $71 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "221", "answer_3": "257", "answer_0": "238", "answer_1": "254"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, November 10, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.300", "p3": "0.222", "p2": "0.700", "p1": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $302. A fee of $146 is charged, then a deposit of $86 arrives. A second fee of $73 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "187", "answer_0": "169", "answer_2": "159", "answer_1": "198"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Friday, March 06, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.500", "p3": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $595 per year. It increases by 5% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "631.75", "answer_2": "565.25", "answer_0": "603.75", "answer_1": "624.75"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1982. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "44", "age_1": "42", "age_0": "43", "age_3": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed February 11, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 16", "date_3": "February 16", "date_2": "February 14", "date_1": "February 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "21", "answer_3": "18", "answer_0": "36", "answer_2": "18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1815 per year. It increases by 20% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1452", "answer_3": "2133", "answer_1": "2178", "answer_0": "2159"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.273", "p1": "0.273", "p0": "0.727", "p3": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed August 24, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 06", "date_3": "September 04", "date_0": "September 04", "date_1": "September 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1012 per year. It increases by 25% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1245", "answer_1": "1265", "answer_3": "1235", "answer_2": "759"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = $364; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.364", "p2": "0.364", "p3": "0.600", "p0": "0.636"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1969. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "59", "age_0": "57", "age_2": "58", "age_1": "56"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.250", "p1": "0.500", "p0": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1981. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "47", "age_2": "46", "age_1": "44", "age_0": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed July 21, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "August 02", "date_3": "August 02", "date_0": "August 04", "date_1": "July 31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $245. A fee of $32 is charged, then a deposit of $90 arrives. A second fee of $16 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "275", "answer_0": "287", "answer_2": "313", "answer_3": "281"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=607, unit_price=875, discount_pct=0, region=EU, sku_total=1482 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "53112", "v3": "530518", "v1": "1482", "v0": "531125"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 3 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 09:30 to 12:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h30", "dur_1": "3h00", "dur_2": "3h00", "dur_3": "2h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=698, unit_price=762, discount_pct=0, region=EU, sku_total=1460 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "531876", "v1": "1460", "v2": "53187", "v3": "531178"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = open; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_1": "27", "answer_0": "42", "answer_3": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 560:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p0": "0.200", "p3": "0.050", "p1": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1747 per year. It increases by 15% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1966.05", "answer_2": "1484.95", "answer_0": "1984.05", "answer_1": "2009.05"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 973:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.800", "p0": "0.200", "p1": "0.800", "p3": "0.111"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.167", "p0": "0.333", "p1": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=184, unit_price=436, discount_pct=0, region=EU, sku_total=620 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "8022", "v0": "80224", "v1": "620", "v3": "80408"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p2": "0.050", "p3": "0.050", "p0": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $812 per year. It increases by 15% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "886.8", "answer_2": "690.2", "answer_1": "933.8", "answer_3": "896.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $33k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $542. A fee of $191 is charged, then a deposit of $86 arrives. A second fee of $95 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "322", "answer_3": "353", "answer_2": "335", "answer_0": "342"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $975 per year. It increases by 15% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1103.25", "answer_3": "1116.25", "answer_1": "1121.25", "answer_2": "828.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = 2026-12-20; priority = high.\n\nClaim to verify: \"the date is 2026-12-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 7 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed March 04, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 15", "date_1": "March 13", "date_2": "March 15", "date_0": "March 17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1963. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "62", "age_0": "63", "age_3": "65", "age_2": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.455", "p0": "0.500", "p1": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.167", "p1": "0.250", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 219 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "296", "answer_3": "312", "answer_2": "321", "answer_0": "315"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, April 28, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, November 03, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = 2026-04-19; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 357 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "112", "answer_2": "123", "answer_1": "118", "answer_3": "105"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = 2026-06-26; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p1": "0.625", "p0": "0.500", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = $314; priority = low.\n\nClaim to verify: \"the status is $314\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=842, unit_price=528, discount_pct=0, region=EU, sku_total=1370 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1370", "v3": "443734", "v2": "44457", "v0": "444576"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1976. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "47", "age_3": "46", "age_0": "48", "age_2": "49"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, September 22, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = 2026-09-28; priority = high.\n\nClaim to verify: \"the date is 2026-09-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 09:30 to 20:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "10h45", "dur_3": "9h45", "dur_2": "10h15", "dur_0": "10h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1981. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "46", "age_0": "45", "age_1": "44", "age_3": "47"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.667", "p3": "0.167", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.111", "p1": "0.800", "p2": "0.800", "p0": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 5 grams of salt per serving and serves 5 people. A shopper bought a 25 gram box and a 250 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p2": "0.250", "p1": "0.500", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 3 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $8k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $593. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 319 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "63", "answer_3": "80", "answer_1": "53", "answer_2": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "9", "answer_1": "18", "answer_2": "15", "answer_0": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed March 01, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 10", "date_3": "March 12", "date_0": "March 12", "date_1": "March 10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 255 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "240", "answer_2": "259", "answer_1": "249", "answer_3": "224"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1984. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "43", "age_1": "41", "age_0": "42", "age_3": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed July 07, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 20", "date_1": "July 16", "date_2": "July 18", "date_3": "July 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_3": "18", "answer_2": "30", "answer_1": "33"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1983. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "43", "age_1": "41", "age_0": "42", "age_3": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = pending; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = $255; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "49", "answer_1": "37", "answer_3": "14", "answer_2": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $366. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 11:15 to 18:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "6h15", "dur_1": "6h45", "dur_2": "6h30", "dur_3": "5h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=913, unit_price=144, discount_pct=0, region=EU, sku_total=1057 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "13147", "v1": "1057", "v0": "131472", "v3": "132385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 370:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = 2026-11-08; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 58 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "107", "answer_0": "120", "answer_2": "128", "answer_1": "109"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.400", "p1": "0.500", "p0": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p1": "0.667", "p0": "0.333", "p3": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = $590; priority = high.\n\nClaim to verify: \"the amount is $590\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $839. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1977. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "49", "age_1": "48", "age_2": "50", "age_3": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.444", "p3": "0.375", "p1": "0.556", "p2": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=996, unit_price=886, discount_pct=0, region=EU, sku_total=1882 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1882", "v2": "88245", "v0": "882456", "v3": "881460"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $345. A fee of $175 is charged, then a deposit of $77 arrives. A second fee of $87 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "155", "answer_1": "166", "answer_0": "160", "answer_2": "166"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, March 09, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 442:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1988. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "38", "age_2": "39", "age_1": "37", "age_3": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1207 per year. It increases by 25% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "905.25", "answer_0": "1485.75", "answer_3": "1446.75", "answer_1": "1508.75"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=925, unit_price=105, discount_pct=0, region=EU, sku_total=1030 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "98050", "v1": "1030", "v0": "97125", "v2": "9712"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.545", "p3": "0.400", "p2": "0.545", "p0": "0.455"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 383 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "172", "answer_2": "173", "answer_3": "201", "answer_0": "192"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed March 20, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 25", "date_3": "March 27", "date_0": "March 27", "date_1": "March 25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.100", "p0": "0.150", "p1": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "16", "answer_1": "28", "answer_2": "24", "answer_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Friday, February 20, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 3 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $354. A fee of $51 is charged, then a deposit of $65 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "327", "answer_2": "365", "answer_0": "343", "answer_3": "346"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=745, unit_price=982, discount_pct=0, region=EU, sku_total=1727 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1727", "v2": "73159", "v0": "731590", "v3": "732335"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = 2026-10-05; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=437, unit_price=369, discount_pct=0, region=EU, sku_total=806 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "16125", "v3": "160816", "v0": "161253", "v1": "806"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 159 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "240", "answer_3": "254", "answer_1": "246", "answer_2": "232"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1619 per year. It increases by 20% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1942.8", "answer_2": "1295.2", "answer_0": "1916.8", "answer_3": "1950.8"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=373, unit_price=131, discount_pct=0, region=EU, sku_total=504 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "48863", "v3": "49236", "v2": "4886", "v1": "504"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 140 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "22", "answer_1": "30", "answer_3": "20", "answer_0": "24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "54", "answer_1": "19", "answer_3": "42", "answer_2": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 242 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "94", "answer_0": "96", "answer_2": "81", "answer_3": "86"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An account starts at $862. A fee of $67 is charged, then a deposit of $21 arrives. A second fee of $33 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "783", "answer_2": "775", "answer_3": "759", "answer_1": "768"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 20:00 to 22:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "1h30", "dur_1": "2h30", "dur_2": "2h15", "dur_0": "1h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.545", "p2": "0.417", "p0": "0.583"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1969. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "58", "age_1": "56", "age_3": "59", "age_0": "57"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = 2026-02-01; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $826. A fee of $111 is charged, then a deposit of $45 arrives. A second fee of $55 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "713", "answer_1": "690", "answer_2": "680", "answer_0": "705"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 102 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "59", "answer_0": "54", "answer_3": "70", "answer_2": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A shift runs from 17:00 to 21:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "4h00", "dur_3": "3h00", "dur_2": "3h30", "dur_0": "3h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 240 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_1": "89", "answer_3": "81", "answer_2": "87"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = $717; priority = high.\n\nClaim to verify: \"the status is $717\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, August 17, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 08:00 to 19:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "10h45", "dur_3": "10h15", "dur_1": "11h15", "dur_0": "10h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Sunday, July 05, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed August 24, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "August 31", "date_3": "August 31", "date_2": "September 02", "date_1": "August 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $227. A fee of $35 is charged, then a deposit of $66 arrives. A second fee of $17 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "241", "answer_1": "250", "answer_2": "251", "answer_3": "253"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 08:00 to 08:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "0h30", "dur_2": "0h15", "dur_0": "-1h45", "dur_3": "-1h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1959. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "65", "age_1": "64", "age_3": "67", "age_2": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 3 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, April 27, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.385", "p1": "0.615", "p3": "0.333", "p2": "0.615"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 06:45 to 19:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "12h30", "dur_2": "12h45", "dur_1": "12h45", "dur_3": "11h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = open; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 747:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=780, unit_price=193, discount_pct=0, region=EU, sku_total=973 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "15054", "v3": "149760", "v0": "150540", "v1": "973"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p3": "0.250", "p0": "0.333", "p2": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed March 11, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 19", "date_3": "March 19", "date_2": "March 17", "date_1": "March 17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=724, unit_price=906, discount_pct=0, region=EU, sku_total=1630 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "65594", "v1": "1630", "v0": "655944", "v3": "655220"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1986. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "38", "age_2": "39", "age_1": "37", "age_3": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1990. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "35", "age_3": "34", "age_0": "36", "age_2": "37"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p2": "0.615", "p1": "0.615", "p0": "0.385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p0": "0.429", "p1": "0.571", "p2": "0.571"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 3 grams of salt per serving and serves 12 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=414, unit_price=175, discount_pct=0, region=EU, sku_total=589 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "7245", "v1": "589", "v0": "72450", "v3": "72036"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.167", "p2": "0.083", "p1": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: date = 2026-06-20; priority = low.\n\nClaim to verify: \"the date is 2026-06-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed January 19, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 27", "date_1": "January 27", "date_3": "January 29", "date_0": "January 29"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = $659; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 9 grams of salt per serving and serves 12 people. A shopper bought a 108 gram box and a 1080 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 14:45 to 22:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "7h30", "dur_1": "7h15", "dur_0": "7h00", "dur_3": "6h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = 2026-07-20; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.364", "p1": "0.364", "p3": "0.600", "p0": "0.636"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, February 09, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 09:45 to 21:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "11h00", "dur_1": "11h15", "dur_2": "11h30", "dur_3": "10h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 759:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = 2026-02-15; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p0": "0.500", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.455", "p0": "0.545", "p1": "0.455"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 351 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "68", "answer_1": "75", "answer_0": "60", "answer_3": "71"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 13:00 to 16:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h45", "dur_2": "3h15", "dur_1": "3h30", "dur_3": "2h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $287. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 132 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "30", "answer_1": "42", "answer_3": "39", "answer_2": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p1": "0.200", "p2": "0.100", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 280 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "325", "answer_3": "320", "answer_2": "347", "answer_0": "336"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = $912; priority = high.\n\nClaim to verify: \"the owner is $912\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "24", "answer_0": "36", "answer_3": "18", "answer_2": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed June 06, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 14", "date_1": "June 12", "date_2": "June 13", "date_0": "June 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $508. A fee of $120 is charged, then a deposit of $75 arrives. A second fee of $60 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "395", "answer_2": "427", "answer_3": "379", "answer_0": "403"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = open; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.556", "p2": "0.444", "p1": "0.444"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = 2026-06-28; priority = high.\n\nClaim to verify: \"the amount is 2026-06-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, January 13, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "10", "answer_3": "8", "answer_0": "16", "answer_2": "8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = 2026-03-08; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1976. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "47", "age_2": "49", "age_0": "48", "age_3": "50"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = 2026-05-15; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = pending; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.583", "p1": "0.385", "p2": "0.385", "p0": "0.615"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1978. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "47", "age_3": "44", "age_0": "46", "age_1": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = $841; priority = low.\n\nClaim to verify: \"the status is $841\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 115 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "53", "answer_3": "76", "answer_2": "58", "answer_0": "72"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, February 03, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed June 19, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "June 25", "date_3": "June 25", "date_1": "June 23", "date_2": "June 23"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.778", "p0": "0.222", "p1": "0.778", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = 2026-10-02; priority = low.\n\nClaim to verify: \"the amount is 2026-10-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed November 01, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 15", "date_0": "November 13", "date_1": "November 11", "date_3": "November 13"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $63. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "42", "answer_3": "21", "answer_1": "45", "answer_0": "63"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $377 per year. It increases by 25% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "474.25", "answer_2": "282.75", "answer_0": "459.25", "answer_1": "471.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 514:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $568 per year. It increases by 5% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "596.4", "answer_2": "539.6", "answer_0": "574.4", "answer_3": "598.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 11 people. A shopper bought a 99 gram box and a 990 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 601:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = 2026-04-05; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = $151; priority = low.\n\nClaim to verify: \"the amount is $151\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $489. A fee of $95 is charged, then a deposit of $86 arrives. A second fee of $47 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "454", "answer_2": "460", "answer_3": "461", "answer_0": "433"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 330 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "62", "answer_1": "83", "answer_2": "93", "answer_0": "75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1552 per year. It increases by 20% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1241.6", "answer_0": "1835.4", "answer_3": "1826.4", "answer_1": "1862.4"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=239, unit_price=389, discount_pct=0, region=EU, sku_total=628 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "628", "v2": "9297", "v0": "92971", "v3": "93210"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1990. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "34", "age_2": "35", "age_3": "36", "age_1": "33"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.625", "p0": "0.500", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: date = $212; priority = low.\n\nClaim to verify: \"the date is $212\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 08:15 to 11:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "3h15", "dur_1": "3h30", "dur_3": "2h30", "dur_2": "3h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: date = $131; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 465:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $510. A fee of $184 is charged, then a deposit of $47 arrives. A second fee of $92 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "306", "answer_0": "281", "answer_1": "301", "answer_2": "304"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "48", "answer_0": "64", "answer_1": "50", "answer_3": "16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = $792; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.385", "p2": "0.615", "p3": "0.333", "p1": "0.615"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = $710; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.667", "p0": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 07:15 to 21:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "14h30", "dur_3": "13h30", "dur_0": "13h45", "dur_2": "14h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p3": "0.462", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = $432; priority = low.\n\nClaim to verify: \"the date is $432\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = $663; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = $225; priority = high.\n\nClaim to verify: \"the date is $225\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 74 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "41", "answer_0": "54", "answer_3": "62", "answer_2": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 3 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "8", "answer_3": "24", "answer_0": "32", "answer_1": "14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed November 19, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 23", "date_1": "November 23", "date_0": "November 25", "date_3": "November 25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = 2026-02-21; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1961. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "67", "age_0": "65", "age_2": "66", "age_1": "64"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 07:00 to 20:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "12h30", "dur_2": "13h30", "dur_0": "13h15", "dur_1": "13h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = $431; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Saturday, October 10, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p1": "0.500", "p3": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1983. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "43", "age_3": "44", "age_1": "41", "age_0": "42"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 93 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_1": "44", "answer_2": "45", "answer_0": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=836, unit_price=574, discount_pct=0, region=EU, sku_total=1410 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "480700", "v0": "479864", "v2": "47986", "v1": "1410"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p1": "0.667", "p3": "0.200", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.286", "p3": "0.667", "p0": "0.714", "p2": "0.286"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 226 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "84", "answer_3": "70", "answer_1": "70", "answer_2": "77"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 227 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "254", "answer_3": "230", "answer_1": "256", "answer_0": "245"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed October 23, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 03", "date_3": "November 01", "date_2": "November 01", "date_1": "October 30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed July 15, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 26", "date_3": "July 24", "date_1": "July 22", "date_0": "July 24"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.083", "p1": "0.417", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_1": "25", "answer_2": "18", "answer_0": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1965. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "60", "age_2": "62", "age_3": "59", "age_0": "61"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "28", "answer_1": "15", "answer_0": "36", "answer_2": "8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 293 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "73", "answer_0": "75", "answer_1": "71", "answer_2": "87"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "37", "answer_3": "14", "answer_0": "49", "answer_2": "35"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p3": "0.125", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 598:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = 2026-08-23; priority = high.\n\nClaim to verify: \"the owner is 2026-08-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=198, unit_price=369, discount_pct=0, region=EU, sku_total=567 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "73062", "v3": "73260", "v1": "567", "v2": "7306"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = $70; priority = high.\n\nClaim to verify: \"the date is $70\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.727", "p2": "0.273", "p1": "0.273", "p3": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "18", "answer_2": "15", "answer_3": "9", "answer_0": "24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p3": "0.083", "p1": "0.250", "p0": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p1": "0.150", "p2": "0.050", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 114 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "200", "answer_3": "188", "answer_0": "180", "answer_2": "162"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = $36; priority = low.\n\nClaim to verify: \"the owner is $36\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = $105; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed May 18, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 27", "date_0": "May 27", "date_1": "May 25", "date_2": "May 25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 07:15 to 15:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "7h45", "dur_2": "8h15", "dur_3": "7h15", "dur_1": "8h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 7 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 261:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $757. A fee of $95 is charged, then a deposit of $70 arrives. A second fee of $47 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "685", "answer_1": "712", "answer_3": "707", "answer_2": "708"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p2": "0.600", "p3": "0.333", "p1": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.083", "p0": "0.250", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 275:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 284:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1964. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "61", "age_3": "60", "age_0": "62", "age_2": "63"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 313 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_2": "26", "answer_1": "44", "answer_3": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $766. A fee of $89 is charged, then a deposit of $35 arrives. A second fee of $44 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "654", "answer_2": "653", "answer_0": "668", "answer_1": "677"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 598:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, May 11, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1989. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "35", "age_2": "38", "age_0": "37", "age_1": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed July 01, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 06", "date_3": "July 06", "date_2": "July 04", "date_1": "July 04"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.444", "p2": "0.444", "p0": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.533", "p0": "0.467", "p3": "0.429", "p1": "0.533"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An account starts at $517. A fee of $107 is charged, then a deposit of $76 arrives. A second fee of $53 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "405", "answer_0": "433", "answer_1": "458", "answer_3": "416"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = $154; priority = high.\n\nClaim to verify: \"the owner is $154\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.375", "p3": "0.125", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed January 17, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 28", "date_1": "January 25", "date_2": "January 30", "date_3": "January 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.333", "p0": "0.500", "p3": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A shift runs from 08:45 to 11:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "2h15", "dur_1": "2h45", "dur_0": "2h00", "dur_3": "1h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 7 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "10", "answer_0": "40", "answer_1": "16", "answer_3": "30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1316 per year. It increases by 25% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "987", "answer_3": "1584", "answer_0": "1624", "answer_1": "1645"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1969. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "55", "age_1": "54", "age_2": "56", "age_3": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = 2026-09-13; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=181, unit_price=485, discount_pct=0, region=EU, sku_total=666 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "8778", "v3": "87966", "v1": "666", "v0": "87785"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 215 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "181", "answer_2": "178", "answer_0": "196", "answer_1": "194"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1980. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "43", "age_3": "46", "age_0": "44", "age_2": "45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = 2026-04-01; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p3": "0.167", "p2": "0.500", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1970. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "54", "age_0": "55", "age_3": "53", "age_2": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $432 per year. It increases by 5% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "410.4", "answer_3": "442.6", "answer_1": "453.6", "answer_0": "435.6"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=511, unit_price=565, discount_pct=0, region=EU, sku_total=1076 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "288204", "v1": "1076", "v2": "28871", "v0": "288715"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 396 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "186", "answer_0": "180", "answer_2": "169", "answer_3": "163"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.375", "p0": "0.250", "p2": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 140 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_3": "72", "answer_2": "95", "answer_1": "77"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = $977; priority = low.\n\nClaim to verify: \"the owner is $977\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = 2026-10-11; priority = low.\n\nClaim to verify: \"the amount is 2026-10-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1958. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "65", "age_0": "66", "age_2": "67", "age_3": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "22", "answer_2": "20", "answer_3": "8", "answer_0": "28"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $16k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p2": "0.125", "p0": "0.250", "p3": "0.125"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $954 per year. It increases by 5% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "906.3", "answer_3": "934.7", "answer_1": "1001.7", "answer_0": "955.7"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $352. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.462", "p2": "0.462", "p0": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 308 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "83", "answer_0": "96", "answer_3": "87", "answer_2": "81"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "33", "answer_1": "19", "answer_3": "21", "answer_2": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Monday, February 16, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = $319; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = 2026-06-12; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, April 09, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 293 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "216", "answer_0": "210", "answer_3": "190", "answer_2": "221"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 183 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "180", "answer_2": "177", "answer_1": "165", "answer_3": "192"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "38", "answer_0": "56", "answer_3": "21", "answer_2": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1969. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "58", "age_1": "55", "age_2": "57", "age_0": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.333", "p3": "0.167", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p3": "0.636", "p1": "0.333", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.100", "p1": "0.150", "p2": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=550, unit_price=890, discount_pct=0, region=EU, sku_total=1440 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "489500", "v3": "490050", "v1": "1440", "v2": "48950"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1972. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "52", "age_0": "53", "age_2": "54", "age_3": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 506:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = $602; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "32", "answer_0": "72", "answer_1": "44", "answer_2": "40"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 8 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "37", "answer_3": "40", "answer_0": "72", "answer_2": "32"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = $816; priority = low.\n\nClaim to verify: \"the date is $816\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.286", "p1": "0.286", "p3": "0.667", "p0": "0.714"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "64", "answer_1": "36", "answer_2": "32", "answer_3": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 82 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "71", "answer_2": "78", "answer_0": "84", "answer_3": "90"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.083", "p0": "0.167", "p1": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.571", "p3": "0.500", "p1": "0.429", "p2": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p3": "0.050", "p2": "0.050", "p0": "0.100"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = $824; priority = low.\n\nClaim to verify: \"the date is $824\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, November 27, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = $992; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=947, unit_price=218, discount_pct=0, region=EU, sku_total=1165 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "20644", "v0": "206446", "v3": "207393", "v1": "1165"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_0": "27", "answer_1": "15", "answer_2": "9"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=464, unit_price=821, discount_pct=0, region=EU, sku_total=1285 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "380944", "v3": "381408", "v2": "38094", "v1": "1285"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 10:30 to 22:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "11h30", "dur_1": "12h00", "dur_2": "11h45", "dur_3": "11h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 272:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=824, unit_price=753, discount_pct=0, region=EU, sku_total=1577 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "62047", "v0": "620472", "v1": "1577", "v3": "619648"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed January 22, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 07", "date_0": "February 05", "date_3": "February 03", "date_1": "February 01"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 9 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Friday, September 25, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "16", "answer_0": "32", "answer_2": "16", "answer_1": "18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = $531; priority = low.\n\nClaim to verify: \"the owner is $531\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $809. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = closed; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 126 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "44", "answer_2": "53", "answer_0": "48", "answer_3": "51"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = closed; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.385", "p2": "0.385", "p0": "0.615", "p3": "0.583"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1983. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "43", "age_0": "41", "age_1": "40", "age_2": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $34k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed January 19, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 26", "date_3": "January 28", "date_0": "January 28", "date_2": "January 30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "20", "answer_3": "28", "answer_1": "27", "answer_0": "48"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 201:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = 2026-09-19; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.250", "p3": "0.083", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $367. A fee of $91 is charged, then a deposit of $28 arrives. A second fee of $45 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "245", "answer_2": "229", "answer_0": "259", "answer_1": "262"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 7 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1970. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "53", "age_2": "55", "age_0": "54", "age_3": "56"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 386 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "42", "answer_2": "23", "answer_1": "40", "answer_3": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 393 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "143", "answer_3": "132", "answer_0": "150", "answer_1": "145"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.167", "p1": "0.833", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 2 grams of salt per serving and serves 8 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $886. A fee of $172 is charged, then a deposit of $50 arrives. A second fee of $86 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "678", "answer_2": "648", "answer_1": "669", "answer_3": "649"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $15k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=190, unit_price=524, discount_pct=0, region=EU, sku_total=714 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "99560", "v2": "9956", "v3": "99750", "v1": "714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed September 20, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 26", "date_2": "September 26", "date_3": "September 28", "date_0": "September 28"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $34k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $817. A fee of $176 is charged, then a deposit of $75 arrives. A second fee of $88 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "612", "answer_0": "628", "answer_2": "648", "answer_3": "658"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = $888; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.700", "p0": "0.300", "p3": "0.222", "p2": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.333", "p3": "0.083", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = $634; priority = high.\n\nClaim to verify: \"the status is $634\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: owner = $47; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 530:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 7 grams of salt per serving and serves 5 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $641. A fee of $181 is charged, then a deposit of $36 arrives. A second fee of $90 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "406", "answer_1": "377", "answer_3": "432", "answer_2": "387"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_2": "9", "answer_1": "15", "answer_0": "27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_0": "33", "answer_1": "23", "answer_3": "15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed January 08, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 16", "date_0": "January 16", "date_2": "January 14", "date_1": "January 14"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1978. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "48", "age_0": "46", "age_2": "47", "age_1": "45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=780, unit_price=376, discount_pct=0, region=EU, sku_total=1156 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "293280", "v2": "29328", "v1": "1156", "v3": "292500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1989. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "35", "age_3": "34", "age_0": "36", "age_2": "37"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 193 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "104", "answer_0": "84", "answer_2": "96", "answer_1": "81"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.667", "p2": "0.333", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $645. A fee of $188 is charged, then a deposit of $63 arrives. A second fee of $94 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "442", "answer_1": "421", "answer_2": "431", "answer_0": "426"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $280. A fee of $133 is charged, then a deposit of $36 arrives. A second fee of $66 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "97", "answer_3": "138", "answer_0": "117", "answer_1": "111"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1974. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "52", "age_0": "50", "age_1": "49", "age_2": "51"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $266. A fee of $78 is charged, then a deposit of $65 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "244", "answer_3": "194", "answer_1": "229", "answer_0": "214"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, February 18, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.167", "p3": "0.167", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Thursday, May 07, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = $609; priority = low.\n\nClaim to verify: \"the amount is $609\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed December 11, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 18", "date_0": "December 22", "date_3": "December 20", "date_2": "December 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=210, unit_price=749, discount_pct=0, region=EU, sku_total=959 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "959", "v2": "15729", "v3": "157500", "v0": "157290"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 19:00 to 21:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "1h30", "dur_1": "2h00", "dur_2": "1h45", "dur_3": "1h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1316 per year. It increases by 5% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1314.8", "answer_0": "1352.8", "answer_2": "1250.2", "answer_1": "1381.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.467", "p3": "0.500", "p1": "0.467", "p0": "0.533"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = closed; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.167", "p2": "0.500", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.667", "p3": "0.167", "p1": "0.833"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 17:00 to 20:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "2h15", "dur_0": "2h45", "dur_1": "3h15", "dur_2": "3h00"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Friday, November 20, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = $669; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: status = $598; priority = high.\n\nClaim to verify: \"the status is $598\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, July 14, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.300", "p2": "0.300", "p0": "0.700", "p3": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Thursday, April 02, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 323:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1966. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_3": "60", "age_2": "59", "age_1": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $505 per year. It increases by 15% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "429.25", "answer_1": "580.75", "answer_3": "573.75", "answer_0": "559.75"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $498 per year. It increases by 25% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "373.5", "answer_0": "602.5", "answer_3": "574.5", "answer_1": "622.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 201 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_3": "114", "answer_1": "113", "answer_2": "109"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 222 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "42", "answer_3": "44", "answer_2": "33", "answer_1": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $907 per year. It increases by 20% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "725.6", "answer_0": "1074.4", "answer_1": "1088.4", "answer_3": "1052.4"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.333", "p0": "0.250", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 190 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "37", "answer_3": "10", "answer_0": "18", "answer_1": "33"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 07:45 to 14:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "5h45", "dur_3": "5h15", "dur_1": "6h15", "dur_0": "5h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_3": "18", "answer_1": "21", "answer_2": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = 2026-10-07; priority = low.\n\nClaim to verify: \"the owner is 2026-10-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.533", "p0": "0.467", "p2": "0.533", "p3": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1159 per year. It increases by 5% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1205.95", "answer_2": "1101.05", "answer_3": "1198.95", "answer_1": "1216.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 294 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "70", "answer_0": "60", "answer_3": "46", "answer_1": "53"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.625", "p1": "0.625", "p0": "0.375", "p3": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1415 per year. It increases by 10% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1518.5", "answer_1": "1556.5", "answer_2": "1273.5", "answer_0": "1543.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = $695; priority = low.\n\nClaim to verify: \"the status is $695\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.125", "p2": "0.250", "p0": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = 2026-06-06; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = closed; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1900 per year. It increases by 5% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1995", "answer_2": "1805", "answer_0": "1968", "answer_3": "1938"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p3": "0.125", "p2": "0.125", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1961. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "66", "age_0": "65", "age_1": "64", "age_3": "67"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $800. A fee of $79 is charged, then a deposit of $85 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "767", "answer_1": "785", "answer_3": "758", "answer_2": "738"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $13k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $402. A fee of $122 is charged, then a deposit of $35 arrives. A second fee of $61 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "242", "answer_0": "254", "answer_1": "252", "answer_2": "230"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $498. A fee of $116 is charged, then a deposit of $82 arrives. A second fee of $58 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "391", "answer_0": "406", "answer_1": "399", "answer_3": "388"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = open; priority = high.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $40k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $272. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "33", "answer_1": "19", "answer_2": "12", "answer_3": "21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed March 20, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 26", "date_1": "March 24", "date_2": "March 28", "date_3": "March 26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $425. A fee of $129 is charged, then a deposit of $30 arrives. A second fee of $64 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "270", "answer_2": "275", "answer_0": "262", "answer_1": "284"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 319 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "96", "answer_2": "101", "answer_3": "114", "answer_1": "110"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 9 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 711:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 264 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "20", "answer_3": "50", "answer_2": "25", "answer_0": "40"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 248 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "200", "answer_2": "198", "answer_1": "203", "answer_3": "191"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 443:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.667", "p3": "0.167", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 900:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 219 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "169", "answer_3": "166", "answer_1": "155", "answer_0": "162"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $539. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, August 15, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.625", "p3": "0.571", "p1": "0.375", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 09:45 to 13:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h45", "dur_3": "3h00", "dur_0": "3h30", "dur_1": "4h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = $652; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=841, unit_price=539, discount_pct=0, region=EU, sku_total=1380 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "45329", "v0": "453299", "v1": "1380", "v3": "452458"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 6 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.429", "p1": "0.500", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Thursday, November 26, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.167", "p1": "0.333", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "28", "answer_0": "56", "answer_3": "28", "answer_1": "32"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.167", "p3": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $700. A fee of $166 is charged, then a deposit of $85 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "536", "answer_1": "540", "answer_2": "541", "answer_3": "542"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p3": "0.250", "p2": "0.667", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 854:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.533", "p3": "0.429", "p1": "0.533", "p0": "0.467"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $221. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.417", "p2": "0.583", "p3": "0.364", "p1": "0.583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.700", "p1": "0.300", "p2": "0.300", "p3": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.667", "p3": "0.167", "p1": "0.833"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.150", "p3": "0.050", "p0": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "17", "answer_3": "30", "answer_0": "42", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 623:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, July 27, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 82 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "87", "answer_0": "80", "answer_2": "89", "answer_3": "86"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $305. A fee of $65 is charged, then a deposit of $82 arrives. A second fee of $32 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "277", "answer_1": "288", "answer_3": "301", "answer_0": "290"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $21k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.400", "p2": "0.545", "p0": "0.455", "p1": "0.545"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 100 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "54", "answer_3": "52", "answer_1": "56", "answer_2": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p2": "0.050", "p1": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "26", "answer_2": "24", "answer_3": "12", "answer_0": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $363. A fee of $158 is charged, then a deposit of $82 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "186", "answer_2": "178", "answer_1": "187", "answer_0": "208"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 6 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.833", "p0": "0.667", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_1": "28", "answer_3": "15", "answer_2": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: date = 2026-03-25; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1208 per year. It increases by 20% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1432.6", "answer_1": "1449.6", "answer_2": "966.4", "answer_3": "1455.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.714", "p0": "0.286", "p2": "0.714", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p3": "0.050", "p0": "0.150", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $246. A fee of $162 is charged, then a deposit of $46 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "49", "answer_2": "72", "answer_3": "79", "answer_1": "73"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 592:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = 2026-07-10; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.167", "p1": "0.500", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $806 per year. It increases by 20% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "945.2", "answer_2": "644.8", "answer_3": "955.2", "answer_1": "967.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $4k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $262 per year. It increases by 5% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "256.1", "answer_2": "248.9", "answer_0": "251.1", "answer_1": "275.1"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $837. A fee of $190 is charged, then a deposit of $77 arrives. A second fee of $95 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "620", "answer_2": "611", "answer_1": "610", "answer_0": "629"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "20", "answer_0": "40", "answer_2": "20", "answer_1": "24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = ops; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=603, unit_price=633, discount_pct=0, region=EU, sku_total=1236 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "38169", "v1": "1236", "v0": "381699", "v3": "382302"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 356:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.125", "p1": "0.375", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = 2026-11-18; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 155 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "117", "answer_0": "108", "answer_2": "101", "answer_1": "100"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = $441; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 360 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "115", "answer_1": "136", "answer_2": "122", "answer_0": "120"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $20k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = $289; priority = low.\n\nClaim to verify: \"the owner is $289\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $481. A fee of $162 is charged, then a deposit of $48 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "271", "answer_1": "267", "answer_0": "286", "answer_2": "301"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $136. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $765 per year. It increases by 15% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "829.75", "answer_2": "650.25", "answer_1": "879.75", "answer_3": "817.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.545", "p1": "0.455", "p2": "0.455", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 8 grams of salt per serving and serves 12 people. A shopper bought a 96 gram box and a 960 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $583 per year. It increases by 10% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "641.3", "answer_3": "634.3", "answer_0": "603.3", "answer_2": "524.7"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, April 28, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $987 per year. It increases by 25% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1233.75", "answer_3": "1225.75", "answer_2": "740.25", "answer_0": "1216.75"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "32", "answer_1": "37", "answer_0": "72", "answer_3": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 355 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "58", "answer_2": "52", "answer_1": "36", "answer_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $821. A fee of $141 is charged, then a deposit of $43 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "627", "answer_0": "653", "answer_1": "680", "answer_2": "632"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $392. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = $830; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.778", "p1": "0.778", "p3": "0.125", "p0": "0.222"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.500", "p2": "0.375", "p1": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $475. A fee of $115 is charged, then a deposit of $44 arrives. A second fee of $57 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "335", "answer_1": "367", "answer_3": "353", "answer_0": "347"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $521. A fee of $55 is charged, then a deposit of $71 arrives. A second fee of $27 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "505", "answer_3": "507", "answer_2": "505", "answer_0": "510"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed March 01, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 11", "date_1": "March 11", "date_0": "March 13", "date_3": "March 13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.143", "p1": "0.750", "p2": "0.750"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1973. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "50", "age_2": "52", "age_0": "51", "age_3": "49"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, October 17, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $334. A fee of $135 is charged, then a deposit of $71 arrives. A second fee of $67 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "222", "answer_2": "191", "answer_3": "209", "answer_0": "203"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, May 10, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_0": "48", "answer_1": "28", "answer_3": "24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 11:45 to 21:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "8h30", "dur_0": "9h00", "dur_2": "9h30", "dur_1": "9h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.667", "p3": "0.167", "p1": "0.833"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $415. A fee of $199 is charged, then a deposit of $66 arrives. A second fee of $99 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "175", "answer_0": "183", "answer_1": "197", "answer_3": "163"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed June 15, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 26", "date_0": "June 26", "date_1": "June 24", "date_2": "June 28"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "42", "answer_3": "35", "answer_1": "47", "answer_0": "77"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = $694; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.667", "p2": "0.333", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.375", "p3": "0.125", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1969. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "56", "age_1": "54", "age_0": "55", "age_3": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=788, unit_price=314, discount_pct=0, region=EU, sku_total=1102 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1102", "v0": "247432", "v3": "246644", "v2": "24743"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $794. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=922, unit_price=385, discount_pct=0, region=EU, sku_total=1307 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "35497", "v1": "1307", "v0": "354970", "v3": "355892"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.167", "p0": "0.667", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $17k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.273", "p0": "0.727", "p2": "0.273", "p3": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = $266; priority = low.\n\nClaim to verify: \"the amount is $266\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 249 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "144", "answer_2": "125", "answer_1": "155", "answer_3": "159"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed October 28, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 05", "date_3": "November 05", "date_2": "November 03", "date_1": "November 03"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $199. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Monday, January 05, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 232 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "27", "answer_3": "36", "answer_1": "52", "answer_0": "42"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 580:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.300", "p2": "0.636", "p1": "0.636", "p0": "0.364"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p2": "0.100", "p3": "0.050", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.250", "p2": "0.083", "p0": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed August 15, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 25", "date_2": "August 28", "date_0": "August 26", "date_1": "August 23"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1946 per year. It increases by 20% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1556.8", "answer_0": "2316.2", "answer_1": "2335.2", "answer_3": "2292.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1178 per year. It increases by 10% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1295.8", "answer_0": "1277.8", "answer_3": "1267.8", "answer_2": "1060.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $328. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $40k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $8k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 9 grams of salt per serving and serves 3 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = 2026-01-04; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p2": "0.100", "p1": "0.200", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = 2026-06-06; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = 2026-10-15; priority = high.\n\nClaim to verify: \"the owner is 2026-10-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed December 05, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 09", "date_3": "December 10", "date_2": "December 07", "date_1": "December 08"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $354. A fee of $136 is charged, then a deposit of $57 arrives. A second fee of $68 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "195", "answer_1": "192", "answer_2": "211", "answer_0": "207"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = 2026-10-11; priority = high.\n\nClaim to verify: \"the amount is 2026-10-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = $467; priority = low.\n\nClaim to verify: \"the amount is $467\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "21", "answer_2": "15", "answer_3": "6", "answer_1": "17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.600", "p0": "0.667", "p1": "0.333", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.750", "p1": "0.222", "p0": "0.778", "p2": "0.222"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 09:30 to 18:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "8h30", "dur_3": "7h30", "dur_2": "8h30", "dur_0": "8h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 9 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "6", "answer_0": "21", "answer_1": "17", "answer_2": "15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = 2026-09-21; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.500", "p2": "0.375", "p1": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 169 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "242", "answer_1": "234", "answer_3": "222", "answer_0": "224"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1971. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "57", "age_0": "55", "age_2": "56", "age_1": "54"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 12 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p0": "0.250", "p1": "0.375", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 178 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "79", "answer_0": "60", "answer_1": "58", "answer_3": "72"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1940 per year. It increases by 10% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1746", "answer_0": "2107", "answer_3": "2120", "answer_1": "2134"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, February 16, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 334:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $5k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.167", "p2": "0.500", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_1": "21", "answer_2": "16", "answer_3": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.667", "p0": "0.714", "p1": "0.286", "p2": "0.286"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 13:30 to 18:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "3h45", "dur_0": "4h00", "dur_2": "4h15", "dur_1": "4h45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_3": "28", "answer_0": "52", "answer_1": "31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p2": "0.571", "p0": "0.429", "p3": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed October 18, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "October 21", "date_2": "October 19", "date_3": "October 23", "date_1": "October 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.167", "p0": "0.667", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=717, unit_price=130, discount_pct=0, region=EU, sku_total=847 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "847", "v3": "93927", "v2": "9321", "v0": "93210"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, November 13, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.778", "p0": "0.222", "p2": "0.778"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 06:00 to 19:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "12h30", "dur_2": "13h00", "dur_1": "13h15", "dur_3": "12h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed November 26, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 01", "date_1": "November 29", "date_3": "December 01", "date_2": "December 03"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.333", "p3": "0.083", "p1": "0.417"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 595:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $305. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 694:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.667", "p2": "0.300", "p0": "0.700", "p1": "0.300"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=917, unit_price=742, discount_pct=0, region=EU, sku_total=1659 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "680414", "v3": "679497", "v2": "68041", "v1": "1659"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 399 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "45", "answer_0": "28", "answer_1": "22", "answer_2": "37"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, April 11, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.500", "p1": "0.833", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 9 grams of salt per serving and serves 6 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 405:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = $800; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 13:30 to 21:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "7h45", "dur_2": "8h15", "dur_3": "7h00", "dur_1": "8h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $162. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $764. A fee of $176 is charged, then a deposit of $27 arrives. A second fee of $88 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "543", "answer_2": "503", "answer_1": "497", "answer_0": "527"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.417", "p2": "0.538", "p1": "0.538", "p0": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed October 08, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 15", "date_0": "October 19", "date_3": "October 17", "date_2": "October 17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $246. A fee of $182 is charged, then a deposit of $65 arrives. A second fee of $91 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "23", "answer_1": "31", "answer_2": "34", "answer_0": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 285 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "321", "answer_3": "324", "answer_0": "336", "answer_1": "320"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = 2026-04-26; priority = high.\n\nClaim to verify: \"the owner is 2026-04-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.467", "p0": "0.500", "p2": "0.500", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 12 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p0": "0.250", "p3": "0.125", "p1": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=614, unit_price=817, discount_pct=0, region=EU, sku_total=1431 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1431", "v3": "502252", "v0": "501638", "v2": "50163"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.125", "p3": "0.125", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $593; priority = low.\n\nClaim to verify: \"the owner is $593\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1266 per year. It increases by 15% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1383.9", "answer_2": "1076.1", "answer_1": "1455.9", "answer_0": "1418.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.200", "p2": "0.727", "p0": "0.273", "p1": "0.727"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = 2026-02-14; priority = low.\n\nClaim to verify: \"the status is 2026-02-14\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed February 28, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 12", "date_1": "March 09", "date_2": "March 10", "date_3": "March 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $430. A fee of $120 is charged, then a deposit of $62 arrives. A second fee of $60 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "331", "answer_3": "282", "answer_1": "327", "answer_0": "312"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = $741; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $537. A fee of $189 is charged, then a deposit of $25 arrives. A second fee of $94 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "302", "answer_1": "293", "answer_3": "259", "answer_0": "279"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $679 per year. It increases by 25% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "509.25", "answer_0": "838.75", "answer_1": "848.75", "answer_3": "818.75"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 3 grams of salt per serving and serves 5 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p0": "0.667", "p2": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, December 07, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Saturday, August 08, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=479, unit_price=834, discount_pct=0, region=EU, sku_total=1313 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "39948", "v0": "399486", "v3": "399007", "v1": "1313"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 796:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 217:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = 2026-05-04; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = $929; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=490, unit_price=732, discount_pct=0, region=EU, sku_total=1222 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "358190", "v0": "358680", "v1": "1222", "v2": "35868"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 4 grams of salt per serving and serves 3 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 06:45 to 20:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "13h15", "dur_2": "13h30", "dur_0": "13h00", "dur_3": "12h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p3": "0.462", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=107, unit_price=976, discount_pct=0, region=EU, sku_total=1083 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "104432", "v1": "1083", "v3": "104325", "v2": "10443"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1970. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "54", "age_3": "57", "age_2": "56", "age_0": "55"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed November 16, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 25", "date_2": "November 27", "date_0": "November 25", "date_1": "November 23"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Wednesday, June 03, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $10k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 352 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "427", "answer_0": "432", "answer_3": "447", "answer_1": "448"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, October 08, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=989, unit_price=188, discount_pct=0, region=EU, sku_total=1177 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "186921", "v0": "185932", "v2": "18593", "v1": "1177"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, November 13, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.333", "p1": "0.333", "p3": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 08:45 to 10:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h15", "dur_1": "1h45", "dur_3": "0h45", "dur_0": "1h00"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = 2026-02-14; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=356, unit_price=576, discount_pct=0, region=EU, sku_total=932 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "20505", "v1": "932", "v3": "205412", "v0": "205056"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 191 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "105", "answer_3": "75", "answer_0": "90", "answer_1": "104"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1979. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "47", "age_0": "46", "age_1": "45", "age_3": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $640. A fee of $139 is charged, then a deposit of $82 arrives. A second fee of $69 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "497", "answer_3": "517", "answer_2": "486", "answer_0": "514"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, October 11, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $298. A fee of $101 is charged, then a deposit of $65 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "235", "answer_1": "196", "answer_0": "212", "answer_2": "241"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1314 per year. It increases by 15% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1116.9", "answer_0": "1480.1", "answer_3": "1446.1", "answer_1": "1511.1"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1974. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "54", "age_1": "51", "age_2": "53", "age_0": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 62 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "169", "answer_0": "150", "answer_3": "131", "answer_1": "147"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.444", "p2": "0.444", "p0": "0.556", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 218:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 09:30 to 20:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "10h30", "dur_3": "10h00", "dur_1": "11h00", "dur_2": "11h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 928:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $706. A fee of $128 is charged, then a deposit of $56 arrives. A second fee of $64 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "600", "answer_1": "563", "answer_3": "559", "answer_0": "570"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "18", "answer_0": "32", "answer_2": "16", "answer_3": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $203. A fee of $150 is charged, then a deposit of $42 arrives. A second fee of $75 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "37", "answer_3": "12", "answer_0": "20", "answer_1": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = 2026-01-06; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = 2026-07-20; priority = low.\n\nClaim to verify: \"the status is 2026-07-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 9 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 503:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.375", "p1": "0.625", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=955, unit_price=926, discount_pct=0, region=EU, sku_total=1881 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1881", "v3": "885285", "v0": "884330", "v2": "88433"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = $492; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = $150; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $53. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = 2026-10-10; priority = low.\n\nClaim to verify: \"the status is 2026-10-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $273. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.333", "p3": "0.083", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 12 people. A shopper bought a 96 gram box and a 960 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 146 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "157", "answer_1": "175", "answer_2": "150", "answer_0": "160"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=886, unit_price=510, discount_pct=0, region=EU, sku_total=1396 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "451860", "v3": "452746", "v2": "45186", "v1": "1396"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "28", "answer_2": "24", "answer_3": "32", "answer_0": "56"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1984. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "43", "age_1": "41", "age_3": "44", "age_0": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $373 per year. It increases by 10% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "389.3", "answer_1": "410.3", "answer_2": "335.7", "answer_3": "407.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1292 per year. It increases by 15% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1470.8", "answer_3": "1444.8", "answer_1": "1485.8", "answer_2": "1098.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 243:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $519. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 825:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "15", "answer_3": "15", "answer_1": "20", "answer_0": "30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 840:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, November 11, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p2": "0.375", "p3": "0.125", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.600", "p3": "0.333", "p0": "0.400", "p1": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, January 05, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1981. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "44", "age_1": "42", "age_3": "41", "age_0": "43"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=378, unit_price=526, discount_pct=0, region=EU, sku_total=904 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "198828", "v2": "19882", "v3": "198450", "v1": "904"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $341. A fee of $126 is charged, then a deposit of $45 arrives. A second fee of $63 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "199", "answer_3": "202", "answer_1": "223", "answer_0": "197"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=611, unit_price=668, discount_pct=0, region=EU, sku_total=1279 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "40814", "v0": "408148", "v3": "408759", "v1": "1279"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=363, unit_price=481, discount_pct=0, region=EU, sku_total=844 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "174240", "v1": "844", "v0": "174603", "v2": "17460"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 764:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, November 06, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = $772; priority = low.\n\nClaim to verify: \"the date is $772\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 9 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p3": "0.050", "p2": "0.050", "p0": "0.100"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = $382; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.333", "p2": "0.500", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = $520; priority = high.\n\nClaim to verify: \"the date is $520\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = 2026-05-23; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 08:00 to 15:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "7h30", "dur_2": "7h30", "dur_3": "6h30", "dur_0": "7h00"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 06:15 to 14:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "7h15", "dur_0": "7h30", "dur_1": "8h15", "dur_2": "7h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1967. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "61", "age_1": "58", "age_0": "59", "age_2": "60"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1977. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "45", "age_2": "48", "age_0": "47", "age_1": "46"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $496. A fee of $102 is charged, then a deposit of $61 arrives. A second fee of $51 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "425", "answer_1": "396", "answer_3": "377", "answer_0": "404"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Friday, November 06, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = 2026-12-18; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = ops; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_3": "36", "answer_0": "54", "answer_1": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = $116; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $10k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 15:15 to 15:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h30", "dur_1": "0h30", "dur_0": "-1h45", "dur_2": "0h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 228:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, November 24, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.250", "p0": "0.333", "p1": "0.417"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p0": "0.167", "p1": "0.250", "p3": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_2": "14", "answer_0": "56", "answer_1": "20"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1958. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "66", "age_2": "69", "age_1": "67", "age_0": "68"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = 2026-05-11; priority = low.\n\nClaim to verify: \"the status is 2026-05-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.462", "p1": "0.462", "p0": "0.538", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $273 per year. It increases by 5% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "265.65", "answer_1": "286.65", "answer_3": "228.65", "answer_2": "259.35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $252. A fee of $83 is charged, then a deposit of $84 arrives. A second fee of $41 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "228", "answer_1": "200", "answer_2": "199", "answer_0": "212"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, April 04, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.364", "p2": "0.583", "p0": "0.417", "p1": "0.583"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p0": "0.800", "p2": "0.200", "p3": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.455", "p2": "0.545", "p3": "0.400", "p1": "0.545"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 814:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=755, unit_price=525, discount_pct=0, region=EU, sku_total=1280 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1280", "v3": "395620", "v2": "39637", "v0": "396375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = 2026-01-08; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1959. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "68", "age_1": "65", "age_0": "66", "age_2": "67"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = 2026-07-14; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = $822; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 376 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "270", "answer_1": "262", "answer_0": "252", "answer_2": "255"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=815, unit_price=604, discount_pct=0, region=EU, sku_total=1419 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1419", "v3": "491445", "v0": "492260", "v2": "49226"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A shift runs from 16:00 to 22:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "6h00", "dur_0": "5h45", "dur_1": "6h15", "dur_3": "5h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = 2026-05-12; priority = low.\n\nClaim to verify: \"the owner is 2026-05-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 5 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed June 17, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 24", "date_1": "June 20", "date_3": "June 22", "date_0": "June 22"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Saturday, June 06, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.417", "p0": "0.333", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 426:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=594, unit_price=512, discount_pct=0, region=EU, sku_total=1106 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1106", "v3": "303534", "v0": "304128", "v2": "30412"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Record fields: status = pending; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "45", "answer_0": "80", "answer_3": "40", "answer_2": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed July 13, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 22", "date_1": "July 20", "date_0": "July 22", "date_2": "July 20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.545", "p1": "0.455", "p2": "0.455", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 3 grams of salt per serving and serves 7 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = $209; priority = high.\n\nClaim to verify: \"the status is $209\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.455", "p2": "0.455", "p0": "0.545", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $538. A fee of $83 is charged, then a deposit of $75 arrives. A second fee of $41 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "487", "answer_2": "468", "answer_0": "489", "answer_1": "506"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 245 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "21", "answer_2": "44", "answer_0": "40", "answer_1": "42"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = $891; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 286 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "432", "answer_1": "422", "answer_3": "419", "answer_2": "448"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $13k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Friday, July 17, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $322. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 645:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1989. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "34", "age_2": "36", "age_3": "33", "age_0": "35"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p1": "0.250", "p2": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, September 18, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed December 21, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 02", "date_3": "December 31", "date_1": "December 29", "date_0": "December 31"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $435. A fee of $58 is charged, then a deposit of $24 arrives. A second fee of $29 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "399", "answer_2": "359", "answer_0": "372", "answer_1": "384"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $247. A fee of $174 is charged, then a deposit of $52 arrives. A second fee of $87 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "38", "answer_1": "16", "answer_2": "13", "answer_3": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 08:15 to 15:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "7h00", "dur_0": "6h45", "dur_3": "6h00", "dur_2": "7h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1988. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "34", "age_1": "35", "age_2": "37", "age_0": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, October 12, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = 2026-11-26; priority = high.\n\nClaim to verify: \"the owner is 2026-11-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $709. A fee of $140 is charged, then a deposit of $44 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "527", "answer_2": "554", "answer_0": "543", "answer_1": "513"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $675. A fee of $51 is charged, then a deposit of $74 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "673", "answer_1": "662", "answer_2": "643", "answer_3": "655"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 10:30 to 20:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "9h30", "dur_2": "10h00", "dur_1": "10h00", "dur_3": "9h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = 2026-02-04; priority = low.\n\nClaim to verify: \"the date is 2026-02-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 9 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=289, unit_price=660, discount_pct=0, region=EU, sku_total=949 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "949", "v3": "191029", "v2": "19074", "v0": "190740"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, August 24, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "30", "answer_0": "66", "answer_1": "41", "answer_2": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_2": "18", "answer_1": "22", "answer_0": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = 2026-01-26; priority = low.\n\nClaim to verify: \"the amount is 2026-01-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 09:45 to 22:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "11h15", "dur_1": "12h15", "dur_2": "12h00", "dur_0": "11h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.636", "p3": "0.300", "p1": "0.636", "p0": "0.364"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $245. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed January 24, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 05", "date_0": "February 06", "date_2": "February 08", "date_1": "February 03"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 3 grams of salt per serving and serves 7 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, February 18, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.125", "p0": "0.375", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.333", "p3": "0.600", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 601:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1090 per year. It increases by 20% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1308", "answer_3": "1239", "answer_2": "872", "answer_0": "1266"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 362 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "105", "answer_1": "122", "answer_3": "86", "answer_2": "112"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $29k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed October 17, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 27", "date_1": "October 25", "date_0": "October 28", "date_2": "October 30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 5 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $643. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = $768; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $351. A fee of $79 is charged, then a deposit of $41 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "274", "answer_1": "303", "answer_3": "268", "answer_2": "257"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.467", "p1": "0.467", "p3": "0.500", "p0": "0.533"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=296, unit_price=263, discount_pct=0, region=EU, sku_total=559 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "77552", "v2": "7784", "v0": "77848", "v1": "559"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed March 02, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 08", "date_3": "March 10", "date_0": "March 10", "date_2": "March 12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.615", "p3": "0.333", "p1": "0.615", "p0": "0.385"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed November 07, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 16", "date_2": "November 19", "date_0": "November 17", "date_1": "November 14"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 9 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = 2026-11-09; priority = low.\n\nClaim to verify: \"the status is 2026-11-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 11:30 to 15:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "3h15", "dur_1": "3h30", "dur_3": "2h30", "dur_2": "3h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Thursday, October 15, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 967:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $938 per year. It increases by 20% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "750.4", "answer_1": "1125.6", "answer_3": "1098.6", "answer_0": "1080.6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=623, unit_price=310, discount_pct=0, region=EU, sku_total=933 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "933", "v2": "19313", "v0": "193130", "v3": "192507"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 19:45 to 21:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h00", "dur_3": "1h00", "dur_2": "1h45", "dur_0": "1h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 10 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "6", "answer_2": "18", "answer_1": "20", "answer_0": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "25", "answer_1": "17", "answer_2": "15", "answer_3": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.417", "p1": "0.417", "p3": "0.545", "p0": "0.583"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=408, unit_price=297, discount_pct=0, region=EU, sku_total=705 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "705", "v3": "121584", "v0": "121176", "v2": "12117"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A shift runs from 06:45 to 20:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "13h45", "dur_1": "13h30", "dur_3": "12h30", "dur_0": "13h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 212 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "108", "answer_3": "97", "answer_1": "98", "answer_2": "114"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = $613; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $342. A fee of $181 is charged, then a deposit of $20 arrives. A second fee of $90 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "65", "answer_1": "85", "answer_0": "91", "answer_3": "72"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p3": "0.050", "p0": "0.200", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 5 grams of salt per serving and serves 6 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $251. A fee of $122 is charged, then a deposit of $83 arrives. A second fee of $61 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "155", "answer_3": "121", "answer_1": "145", "answer_0": "151"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Wednesday, May 13, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: date = 2026-05-07; priority = high.\n\nClaim to verify: \"the date is 2026-05-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.125", "p1": "0.375", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1975. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "50", "age_3": "51", "age_0": "49", "age_1": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p0": "0.500", "p3": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = $681; priority = low.\n\nClaim to verify: \"the amount is $681\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Thursday, November 12, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 229 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_1": "106", "answer_2": "114", "answer_3": "116"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $22k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1815 per year. It increases by 20% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2155", "answer_3": "2130", "answer_1": "2178", "answer_2": "1452"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = 2026-09-15; priority = high.\n\nClaim to verify: \"the amount is 2026-09-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, April 11, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.417", "p2": "0.250", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Tuesday, April 21, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=798, unit_price=966, discount_pct=0, region=EU, sku_total=1764 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1764", "v0": "770868", "v2": "77086", "v3": "771666"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.083", "p1": "0.250", "p0": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 84 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "200", "answer_1": "208", "answer_3": "195", "answer_2": "193"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: status = $909; priority = low.\n\nClaim to verify: \"the status is $909\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.625", "p0": "0.667", "p2": "0.333", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $521 per year. It increases by 10% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "548.1", "answer_2": "468.9", "answer_1": "573.1", "answer_3": "536.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $852. A fee of $101 is charged, then a deposit of $76 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "777", "answer_1": "751", "answer_2": "804", "answer_3": "807"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "70", "answer_2": "35", "answer_1": "40", "answer_3": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1982. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "44", "age_3": "46", "age_2": "45", "age_1": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=535, unit_price=506, discount_pct=0, region=EU, sku_total=1041 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "270710", "v1": "1041", "v3": "271245", "v2": "27071"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = $590; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.462", "p0": "0.538", "p2": "0.462", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, October 05, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $692. A fee of $121 is charged, then a deposit of $87 arrives. A second fee of $60 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "607", "answer_0": "598", "answer_3": "579", "answer_1": "574"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = $830; priority = high.\n\nClaim to verify: \"the owner is $830\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $35k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "28", "answer_2": "24", "answer_3": "24", "answer_0": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p0": "0.100", "p3": "0.050", "p2": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = $903; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 132 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "138", "answer_1": "122", "answer_2": "114", "answer_0": "120"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $16k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.500", "p1": "0.833", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed April 13, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 16", "date_2": "April 18", "date_3": "April 18", "date_0": "April 16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = 2026-03-05; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=442, unit_price=705, discount_pct=0, region=EU, sku_total=1147 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1147", "v0": "311610", "v2": "31161", "v3": "311168"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=229, unit_price=555, discount_pct=0, region=EU, sku_total=784 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "127095", "v3": "127324", "v1": "784", "v2": "12709"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.778", "p1": "0.778", "p3": "0.125", "p0": "0.222"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p0": "0.667", "p2": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p1": "0.333", "p3": "0.600", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 312 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "49", "answer_0": "36", "answer_2": "33", "answer_3": "26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p0": "0.500", "p2": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 607:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1911 per year. It increases by 20% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1528.8", "answer_1": "2293.2", "answer_3": "2298.2", "answer_0": "2258.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_2": "42", "answer_1": "48", "answer_0": "84"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 251 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "108", "answer_0": "120", "answer_1": "140", "answer_3": "117"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_1": "19", "answer_0": "28", "answer_3": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "28", "answer_0": "40", "answer_2": "24", "answer_3": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed May 01, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 07", "date_2": "May 13", "date_3": "May 09", "date_0": "May 11"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 09:30 to 10:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "0h30", "dur_3": "-1h45", "dur_2": "1h00", "dur_1": "0h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=938, unit_price=799, discount_pct=0, region=EU, sku_total=1737 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "749462", "v2": "74946", "v1": "1737", "v3": "748524"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.250", "p3": "0.083", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 592:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 17:00 to 17:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "0h45", "dur_0": "0h15", "dur_2": "0h45", "dur_3": "-1h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = $597; priority = low.\n\nClaim to verify: \"the date is $597\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p3": "0.050", "p0": "0.150", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $11k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = 2026-01-21; priority = high.\n\nClaim to verify: \"the date is 2026-01-21\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 07:30 to 14:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "6h45", "dur_2": "7h00", "dur_0": "6h30", "dur_3": "5h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=729, unit_price=189, discount_pct=0, region=EU, sku_total=918 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "138510", "v1": "918", "v2": "13778", "v0": "137781"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed July 01, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 10", "date_2": "July 12", "date_0": "July 10", "date_1": "July 08"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.615", "p1": "0.615", "p0": "0.385", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = 2026-08-01; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = 2026-12-08; priority = high.\n\nClaim to verify: \"the date is 2026-12-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "25", "answer_0": "60", "answer_2": "18", "answer_3": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, April 15, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 552:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_3": "18", "answer_1": "24", "answer_0": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.778", "p0": "0.800", "p1": "0.200", "p2": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=289, unit_price=272, discount_pct=0, region=EU, sku_total=561 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "78319", "v2": "7860", "v0": "78608", "v1": "561"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1971 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2069.55", "answer_3": "2063.55", "answer_0": "2037.55", "answer_2": "1872.45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = $84; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, May 14, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=466, unit_price=114, discount_pct=0, region=EU, sku_total=580 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "580", "v0": "53124", "v3": "52658", "v2": "5312"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p2": "0.375", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = $741; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.167", "p2": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.100", "p1": "0.150", "p2": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=532, unit_price=180, discount_pct=0, region=EU, sku_total=712 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "9576", "v3": "95228", "v1": "712", "v0": "95760"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed January 08, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 14", "date_2": "January 12", "date_3": "January 14", "date_1": "January 12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1985. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "38", "age_2": "41", "age_1": "39", "age_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, February 20, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $227. A fee of $38 is charged, then a deposit of $48 arrives. A second fee of $19 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "218", "answer_3": "201", "answer_2": "239", "answer_1": "246"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = 2026-03-05; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1503 per year. It increases by 10% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1629.3", "answer_2": "1352.7", "answer_1": "1653.3", "answer_0": "1610.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = $351; priority = low.\n\nClaim to verify: \"the amount is $351\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.364", "p3": "0.300", "p1": "0.636", "p2": "0.636"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1964. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "62", "age_3": "60", "age_1": "61", "age_2": "63"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = 2026-06-06; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.364", "p1": "0.583", "p0": "0.417", "p2": "0.583"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1072 per year. It increases by 25% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "804", "answer_3": "1327", "answer_1": "1340", "answer_0": "1314"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.455", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.125", "p1": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_3": "24", "answer_1": "40", "answer_2": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $710. A fee of $126 is charged, then a deposit of $89 arrives. A second fee of $63 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "610", "answer_3": "606", "answer_1": "592", "answer_2": "630"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed March 15, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 26", "date_1": "March 22", "date_3": "March 24", "date_0": "March 24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1045 per year. It increases by 15% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1201.75", "answer_3": "1160.75", "answer_0": "1189.75", "answer_2": "888.25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=352, unit_price=884, discount_pct=0, region=EU, sku_total=1236 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "31116", "v3": "311520", "v1": "1236", "v0": "311168"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $16k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=695, unit_price=489, discount_pct=0, region=EU, sku_total=1184 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "339160", "v2": "33985", "v1": "1184", "v0": "339855"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = 2026-08-22; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 6 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Thursday, July 09, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=885, unit_price=158, discount_pct=0, region=EU, sku_total=1043 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1043", "v3": "138945", "v2": "13983", "v0": "139830"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p2": "0.250", "p0": "0.333", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=409, unit_price=386, discount_pct=0, region=EU, sku_total=795 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "157874", "v2": "15787", "v1": "795", "v3": "157465"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p3": "0.125", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 734:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $659. A fee of $149 is charged, then a deposit of $34 arrives. A second fee of $74 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "451", "answer_0": "470", "answer_3": "448", "answer_1": "468"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.083", "p1": "0.333", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "24", "answer_2": "18", "answer_0": "54", "answer_3": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = 2026-02-11; priority = high.\n\nClaim to verify: \"the amount is 2026-02-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 17:30 to 18:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "0h15", "dur_1": "0h45", "dur_3": "-1h45", "dur_0": "0h00"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $610. A fee of $74 is charged, then a deposit of $80 arrives. A second fee of $37 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "579", "answer_1": "594", "answer_2": "592", "answer_3": "566"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = 2026-01-18; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $460. A fee of $69 is charged, then a deposit of $47 arrives. A second fee of $34 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "393", "answer_0": "404", "answer_3": "409", "answer_1": "432"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 7 people. A shopper bought a 14 gram box and a 140 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p0": "0.667", "p3": "0.167", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 348 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "163", "answer_1": "193", "answer_3": "170", "answer_0": "180"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $493 per year. It increases by 5% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "468.35", "answer_3": "487.65", "answer_1": "517.65", "answer_0": "505.65"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.111", "p2": "0.800", "p0": "0.200", "p1": "0.800"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1532 per year. It increases by 15% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1302.2", "answer_1": "1761.8", "answer_0": "1743.8", "answer_3": "1771.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.143", "p2": "0.750", "p0": "0.250", "p1": "0.750"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 985:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $349. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1972. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "55", "age_1": "53", "age_0": "54", "age_3": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 278:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, May 12, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $23k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed September 10, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 17", "date_1": "September 15", "date_3": "September 17", "date_2": "September 19"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $10k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 761:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = pending; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 252:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 613:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.200", "p3": "0.050", "p2": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 11:00 to 15:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "4h00", "dur_2": "4h30", "dur_3": "3h30", "dur_1": "4h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 5 people. A shopper bought a 25 gram box and a 250 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed August 11, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 17", "date_1": "August 15", "date_0": "August 17", "date_2": "August 19"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = 2026-12-24; priority = low.\n\nClaim to verify: \"the date is 2026-12-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1964. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "62", "age_1": "60", "age_3": "59", "age_0": "61"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 3 grams of salt per serving and serves 9 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.222", "p3": "0.125", "p1": "0.778", "p2": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = 2026-04-26; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.467", "p0": "0.533", "p3": "0.500", "p2": "0.467"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1981. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "41", "age_2": "44", "age_0": "43", "age_1": "42"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed July 04, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 15", "date_1": "July 12", "date_2": "July 13", "date_3": "July 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=712, unit_price=419, discount_pct=0, region=EU, sku_total=1131 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "297616", "v1": "1131", "v0": "298328", "v2": "29832"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $14k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $244. A fee of $157 is charged, then a deposit of $48 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "57", "answer_1": "59", "answer_3": "64", "answer_2": "76"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 121 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "54", "answer_3": "62", "answer_0": "72", "answer_1": "70"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = 2026-12-03; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1083 per year. It increases by 10% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1195.3", "answer_1": "1191.3", "answer_0": "1179.3", "answer_2": "974.7"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.333", "p0": "0.500", "p3": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Thursday, August 27, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "33", "answer_0": "48", "answer_2": "30", "answer_3": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 741:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = 2026-07-18; priority = high.\n\nClaim to verify: \"the owner is 2026-07-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = $314; priority = high.\n\nClaim to verify: \"the status is $314\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = 2026-02-18; priority = high.\n\nClaim to verify: \"the status is 2026-02-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1417 per year. It increases by 20% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1673.4", "answer_2": "1133.6", "answer_0": "1678.4", "answer_1": "1700.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, September 15, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 10:30 to 22:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "12h00", "dur_3": "11h00", "dur_0": "11h45", "dur_2": "12h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.385", "p1": "0.571", "p2": "0.571", "p0": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 10:15 to 21:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "10h00", "dur_1": "11h00", "dur_2": "11h15", "dur_0": "10h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.727", "p0": "0.273", "p2": "0.727", "p3": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed January 05, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 14", "date_1": "January 10", "date_3": "January 12", "date_0": "January 12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.750", "p0": "0.250", "p3": "0.143", "p1": "0.750"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.727", "p3": "0.700", "p2": "0.273", "p1": "0.273"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "49", "answer_2": "28", "answer_3": "21", "answer_1": "31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = 2026-12-05; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1978. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "48", "age_1": "47", "age_2": "49", "age_3": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed July 09, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 17", "date_0": "July 17", "date_2": "July 15", "date_1": "July 15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $686. A fee of $79 is charged, then a deposit of $47 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "613", "answer_1": "633", "answer_0": "615", "answer_2": "625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.455", "p2": "0.545", "p1": "0.545", "p3": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.538", "p1": "0.538", "p3": "0.417", "p0": "0.462"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 11 people. A shopper bought a 66 gram box and a 660 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = ops; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.500", "p1": "0.667", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 78 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_3": "65", "answer_2": "66", "answer_1": "80"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 08:30 to 12:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "4h00", "dur_0": "3h45", "dur_2": "4h00", "dur_3": "3h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 758:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 06:00 to 15:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "9h45", "dur_2": "10h00", "dur_3": "8h45", "dur_0": "9h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 380 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "280", "answer_2": "298", "answer_3": "289", "answer_1": "297"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $587. A fee of $59 is charged, then a deposit of $56 arrives. A second fee of $29 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "578", "answer_1": "538", "answer_2": "544", "answer_0": "555"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, March 07, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=347, unit_price=900, discount_pct=0, region=EU, sku_total=1247 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1247", "v0": "312300", "v2": "31230", "v3": "311953"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.556", "p2": "0.556", "p3": "0.375", "p0": "0.444"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1425 per year. It increases by 20% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1140", "answer_0": "1690", "answer_3": "1659", "answer_1": "1710"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Wednesday, February 04, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Wednesday, January 21, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = 2026-05-22; priority = high.\n\nClaim to verify: \"the status is 2026-05-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.167", "p1": "0.250", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = $816; priority = low.\n\nClaim to verify: \"the amount is $816\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1987. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "36", "age_0": "37", "age_2": "38", "age_3": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=527, unit_price=194, discount_pct=0, region=EU, sku_total=721 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "721", "v2": "10223", "v3": "101711", "v0": "102238"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = $198; priority = high.\n\nClaim to verify: \"the status is $198\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_3": "36", "answer_1": "36", "answer_0": "66"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed July 26, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "August 05", "date_3": "August 05", "date_1": "August 03", "date_2": "August 07"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "84", "answer_2": "42", "answer_3": "42", "answer_1": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p2": "0.050", "p1": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p2": "0.600", "p3": "0.250", "p0": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 192 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "43", "answer_0": "45", "answer_3": "38", "answer_1": "32"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = $944; priority = low.\n\nClaim to verify: \"the status is $944\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1990. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "35", "age_3": "36", "age_0": "34", "age_1": "33"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = ops; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 756:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 4 grams of salt per serving and serves 5 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed July 12, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 22", "date_1": "July 20", "date_2": "July 24", "date_3": "July 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Thursday, April 23, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 502:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = 2026-04-05; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = 2026-02-14; priority = low.\n\nClaim to verify: \"the amount is 2026-02-14\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p0": "0.250", "p1": "0.375", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p0": "0.200", "p3": "0.050", "p1": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 798:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, October 01, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=120, unit_price=246, discount_pct=0, region=EU, sku_total=366 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "29520", "v3": "29400", "v1": "366", "v2": "2952"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = open; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1970. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "57", "age_3": "58", "age_0": "56", "age_1": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = $876; priority = high.\n\nClaim to verify: \"the amount is $876\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $827. A fee of $112 is charged, then a deposit of $27 arrives. A second fee of $56 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "686", "answer_1": "703", "answer_2": "678", "answer_3": "683"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = 2026-10-04; priority = low.\n\nClaim to verify: \"the date is 2026-10-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $455 per year. It increases by 15% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "512.25", "answer_1": "523.25", "answer_2": "386.75", "answer_3": "540.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = $305; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1970. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "56", "age_1": "55", "age_3": "54", "age_2": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed April 28, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 09", "date_0": "May 11", "date_3": "May 09", "date_1": "May 07"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = $635; priority = low.\n\nClaim to verify: \"the status is $635\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed November 06, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 19", "date_2": "November 17", "date_3": "November 17", "date_1": "November 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $742. A fee of $112 is charged, then a deposit of $80 arrives. A second fee of $56 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "631", "answer_2": "648", "answer_1": "636", "answer_0": "654"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 338 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "83", "answer_1": "57", "answer_3": "83", "answer_0": "72"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1986. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "39", "age_1": "37", "age_3": "36", "age_0": "38"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed December 13, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 18", "date_0": "December 18", "date_2": "December 16", "date_3": "December 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: status = ops; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.167", "p3": "0.167", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = $562; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 353 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "38", "answer_0": "36", "answer_2": "24", "answer_3": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.125", "p0": "0.375", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=276, unit_price=706, discount_pct=0, region=EU, sku_total=982 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "194856", "v2": "19485", "v1": "982", "v3": "195132"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 3 grams of salt per serving and serves 5 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 938:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = $643; priority = high.\n\nClaim to verify: \"the owner is $643\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "30", "answer_0": "72", "answer_2": "24", "answer_3": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.250", "p1": "0.417", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 06:45 to 07:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h30", "dur_2": "0h30", "dur_1": "0h30", "dur_0": "0h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 233 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "195", "answer_0": "189", "answer_2": "174", "answer_1": "198"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.286", "p0": "0.375", "p1": "0.625", "p2": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.111", "p2": "0.800", "p1": "0.800", "p0": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed February 15, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 20", "date_3": "February 22", "date_2": "February 18", "date_1": "February 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, July 14, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $379 per year. It increases by 5% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "377.95", "answer_1": "397.95", "answer_3": "404.95", "answer_2": "360.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.250", "p2": "0.167", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.429", "p3": "0.385", "p2": "0.571", "p1": "0.571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 602:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = open; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1985. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "37", "age_1": "38", "age_2": "40", "age_0": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "20", "answer_3": "12", "answer_2": "8", "answer_1": "11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed June 28, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 01", "date_3": "July 03", "date_2": "June 29", "date_0": "July 01"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=414, unit_price=185, discount_pct=0, region=EU, sku_total=599 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "76590", "v2": "7659", "v1": "599", "v3": "76176"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, April 08, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.800", "p0": "0.200", "p2": "0.800", "p3": "0.111"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.364", "p2": "0.364", "p3": "0.600", "p0": "0.636"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = open; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $12k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Saturday, February 14, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 10 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed September 17, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 27", "date_3": "September 27", "date_1": "September 25", "date_0": "September 29"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 83 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "50", "answer_1": "36", "answer_0": "54", "answer_3": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = $722; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Friday, August 07, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "30", "answer_0": "50", "answer_2": "25", "answer_3": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=158, unit_price=391, discount_pct=0, region=EU, sku_total=549 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "6177", "v0": "61778", "v1": "549", "v3": "61936"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = open; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 336:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 627:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.667", "p2": "0.333", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.167", "p3": "0.083", "p2": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=665, unit_price=392, discount_pct=0, region=EU, sku_total=1057 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "261345", "v2": "26068", "v1": "1057", "v0": "260680"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.250", "p0": "0.750", "p3": "0.714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1111 per year. It increases by 15% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1244.65", "answer_2": "944.35", "answer_1": "1277.65", "answer_3": "1265.65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.300", "p1": "0.300", "p3": "0.667", "p0": "0.700"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 19:30 to 21:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "2h00", "dur_1": "2h15", "dur_0": "1h30", "dur_3": "1h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 375 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "50", "answer_2": "50", "answer_0": "40", "answer_3": "28"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.083", "p0": "0.333", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $17k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Thursday, May 14, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = $190; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.167", "p1": "0.333", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.333", "p1": "0.333", "p3": "0.636"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=239, unit_price=742, discount_pct=0, region=EU, sku_total=981 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "17733", "v3": "177577", "v1": "981", "v0": "177338"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $511. A fee of $155 is charged, then a deposit of $26 arrives. A second fee of $77 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "289", "answer_0": "305", "answer_2": "277", "answer_3": "295"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "25", "answer_0": "40", "answer_3": "20", "answer_2": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $592. A fee of $44 is charged, then a deposit of $66 arrives. A second fee of $22 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "592", "answer_1": "616", "answer_2": "567", "answer_3": "613"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $631 per year. It increases by 25% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "777.75", "answer_0": "765.75", "answer_1": "788.75", "answer_2": "473.25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 10:45 to 13:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "1h15", "dur_1": "2h15", "dur_0": "1h45", "dur_2": "2h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 08:45 to 20:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "10h15", "dur_0": "10h45", "dur_2": "11h15", "dur_1": "11h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, March 10, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 448:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=338, unit_price=158, discount_pct=0, region=EU, sku_total=496 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "496", "v2": "5340", "v0": "53404", "v3": "53742"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = pending; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $548. A fee of $131 is charged, then a deposit of $24 arrives. A second fee of $65 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "405", "answer_0": "376", "answer_1": "404", "answer_3": "347"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed August 01, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "August 08", "date_0": "August 10", "date_1": "August 07", "date_3": "August 09"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 3 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = pending; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.375", "p2": "0.125", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "70", "answer_1": "34", "answer_2": "28", "answer_3": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.778", "p1": "0.778", "p3": "0.125", "p0": "0.222"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $264. A fee of $122 is charged, then a deposit of $27 arrives. A second fee of $61 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "102", "answer_3": "129", "answer_1": "126", "answer_0": "108"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1764 per year. It increases by 25% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2205", "answer_2": "1323", "answer_3": "2178", "answer_0": "2185"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = 2026-07-05; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=294, unit_price=571, discount_pct=0, region=EU, sku_total=865 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "167874", "v3": "168168", "v2": "16787", "v1": "865"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, February 07, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Monday, August 03, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.333", "p3": "0.625", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_1": "16", "answer_0": "28", "answer_2": "14"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 803:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=744, unit_price=231, discount_pct=0, region=EU, sku_total=975 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "17186", "v0": "171864", "v1": "975", "v3": "172608"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1973. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "54", "age_3": "51", "age_0": "53", "age_1": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.125", "p1": "0.625", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.727", "p3": "0.700", "p1": "0.273", "p2": "0.273"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = $372; priority = low.\n\nClaim to verify: \"the owner is $372\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, February 07, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Monday, February 16, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=356, unit_price=820, discount_pct=0, region=EU, sku_total=1176 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "291564", "v1": "1176", "v2": "29192", "v0": "291920"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = $716; priority = high.\n\nClaim to verify: \"the date is $716\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = $49; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p3": "0.050", "p2": "0.050", "p1": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 367 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "131", "answer_2": "149", "answer_0": "144", "answer_3": "135"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 3 grams of salt per serving and serves 6 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=360, unit_price=292, discount_pct=0, region=EU, sku_total=652 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "652", "v3": "104760", "v0": "105120", "v2": "10512"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 301:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 715:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 08:15 to 18:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "9h30", "dur_1": "10h30", "dur_0": "10h15", "dur_2": "10h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $657. A fee of $160 is charged, then a deposit of $66 arrives. A second fee of $80 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "508", "answer_3": "455", "answer_1": "496", "answer_0": "483"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, May 04, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 368:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1960. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "65", "age_1": "64", "age_2": "66", "age_3": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $559. A fee of $64 is charged, then a deposit of $84 arrives. A second fee of $32 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "547", "answer_2": "571", "answer_1": "556", "answer_3": "569"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.083", "p0": "0.167", "p1": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $248. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = 2026-12-10; priority = high.\n\nClaim to verify: \"the amount is 2026-12-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=421, unit_price=576, discount_pct=0, region=EU, sku_total=997 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "242496", "v2": "24249", "v3": "242075", "v1": "997"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1982. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "43", "age_3": "41", "age_1": "42", "age_2": "44"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=218, unit_price=721, discount_pct=0, region=EU, sku_total=939 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "939", "v0": "157178", "v2": "15717", "v3": "156960"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p0": "0.385", "p2": "0.615", "p1": "0.615"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1163 per year. It increases by 10% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1236.3", "answer_1": "1279.3", "answer_2": "1046.7", "answer_0": "1242.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "30", "answer_3": "36", "answer_0": "60", "answer_2": "24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 12 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = $809; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $30k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=159, unit_price=860, discount_pct=0, region=EU, sku_total=1019 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "13674", "v3": "136581", "v0": "136740", "v1": "1019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $284. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1901 per year. It increases by 15% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2186.15", "answer_2": "1615.85", "answer_3": "2158.15", "answer_0": "2170.15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = $898; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed December 22, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 07", "date_3": "January 03", "date_0": "January 05", "date_1": "January 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = 2026-07-02; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: status = ops; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = $307; priority = low.\n\nClaim to verify: \"the date is $307\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 233 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "99", "answer_1": "89", "answer_2": "113", "answer_0": "96"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 194 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "47", "answer_0": "54", "answer_3": "45", "answer_2": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = 2026-03-18; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $604. A fee of $175 is charged, then a deposit of $80 arrives. A second fee of $87 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "442", "answer_0": "422", "answer_3": "410", "answer_1": "420"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "16", "answer_3": "24", "answer_0": "36", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 12 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1969. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "56", "age_3": "55", "age_0": "57", "age_2": "58"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p1": "0.625", "p3": "0.125", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.667", "p3": "0.167", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, October 06, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $687. A fee of $60 is charged, then a deposit of $79 arrives. A second fee of $30 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "663", "answer_2": "662", "answer_3": "653", "answer_0": "676"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p2": "0.600", "p1": "0.600", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.571", "p1": "0.375", "p0": "0.625", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_2": "20", "answer_0": "55", "answer_1": "27"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = 2026-10-17; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.125", "p2": "0.250", "p0": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1772 per year. It increases by 25% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2167", "answer_2": "1329", "answer_3": "2133", "answer_1": "2215"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $543 per year. It increases by 10% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "535.3", "answer_1": "597.3", "answer_0": "547.3", "answer_2": "488.7"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.455", "p0": "0.545", "p2": "0.455", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 749:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = 2026-02-13; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 7 people. A shopper bought a 14 gram box and a 140 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = $738; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = $289; priority = low.\n\nClaim to verify: \"the date is $289\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 229:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_0": "54", "answer_3": "24", "answer_1": "34"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $17k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = 2026-03-04; priority = low.\n\nClaim to verify: \"the amount is 2026-03-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, January 21, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1986. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "41", "age_0": "40", "age_3": "38", "age_1": "39"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 607:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=253, unit_price=788, discount_pct=0, region=EU, sku_total=1041 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1041", "v0": "199364", "v3": "199617", "v2": "19936"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Monday, July 20, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 5 grams of salt per serving and serves 9 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=440, unit_price=507, discount_pct=0, region=EU, sku_total=947 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "223080", "v2": "22308", "v3": "222640", "v1": "947"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.083", "p0": "0.167", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Tuesday, January 20, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = 2026-03-02; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 689:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed September 22, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 29", "date_1": "September 27", "date_2": "September 27", "date_0": "September 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $221. A fee of $87 is charged, then a deposit of $63 arrives. A second fee of $43 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "160", "answer_2": "162", "answer_0": "154", "answer_3": "181"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $380. A fee of $124 is charged, then a deposit of $62 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "260", "answer_3": "232", "answer_2": "227", "answer_0": "256"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "30", "answer_0": "42", "answer_3": "14", "answer_2": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=871, unit_price=816, discount_pct=0, region=EU, sku_total=1687 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "71073", "v0": "710736", "v1": "1687", "v3": "711607"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.333", "p1": "0.667", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1759 per year. It increases by 25% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2168.75", "answer_1": "2198.75", "answer_2": "1319.25", "answer_3": "2152.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.167", "p1": "0.250", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $608. A fee of $196 is charged, then a deposit of $73 arrives. A second fee of $98 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "387", "answer_1": "371", "answer_3": "411", "answer_2": "402"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Thursday, August 06, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.125", "p2": "0.125", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "21", "answer_0": "45", "answer_2": "15", "answer_3": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1811 per year. It increases by 10% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1971.1", "answer_2": "1629.9", "answer_1": "1992.1", "answer_3": "1988.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 388 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_3": "44", "answer_2": "62", "answer_1": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.462", "p3": "0.500", "p0": "0.538", "p1": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed September 26, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 03", "date_3": "October 05", "date_0": "October 06", "date_2": "October 08"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = $498; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p2": "0.800", "p3": "0.111", "p1": "0.800"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 12 people. A shopper bought a 84 gram box and a 840 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $480. A fee of $81 is charged, then a deposit of $70 arrives. A second fee of $40 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "459", "answer_1": "434", "answer_3": "405", "answer_0": "429"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 312:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 280 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "53", "answer_3": "38", "answer_0": "45", "answer_1": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 07:15 to 17:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "9h15", "dur_1": "10h15", "dur_0": "9h45", "dur_2": "10h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p3": "0.050", "p0": "0.200", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 09:15 to 13:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "4h30", "dur_2": "4h15", "dur_3": "3h30", "dur_0": "3h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $723. A fee of $107 is charged, then a deposit of $30 arrives. A second fee of $53 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "581", "answer_0": "593", "answer_1": "604", "answer_3": "604"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=758, unit_price=583, discount_pct=0, region=EU, sku_total=1341 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1341", "v0": "441914", "v2": "44191", "v3": "442672"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=598, unit_price=664, discount_pct=0, region=EU, sku_total=1262 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "39707", "v3": "397670", "v0": "397072", "v1": "1262"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $694. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.500", "p0": "0.333", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = 2026-06-20; priority = high.\n\nClaim to verify: \"the status is 2026-06-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, May 26, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.714", "p0": "0.286", "p3": "0.167", "p2": "0.714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = 2026-03-25; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 653:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = $336; priority = low.\n\nClaim to verify: \"the date is $336\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p0": "0.167", "p1": "0.250", "p3": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1987. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "37", "age_0": "38", "age_3": "36", "age_2": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.500", "p3": "0.400", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.500", "p1": "0.667", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1979. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "46", "age_3": "43", "age_0": "45", "age_1": "44"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.615", "p0": "0.385", "p1": "0.615", "p3": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = $393; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_2": "12", "answer_1": "15", "answer_3": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 469:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 266:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.143", "p2": "0.750", "p0": "0.250", "p1": "0.750"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1204 per year. It increases by 20% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1433.8", "answer_2": "963.2", "answer_0": "1417.8", "answer_1": "1444.8"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 117 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "203", "answer_2": "220", "answer_0": "210", "answer_1": "212"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.286", "p1": "0.625", "p0": "0.375", "p2": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A shift runs from 06:00 to 21:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "15h30", "dur_3": "14h30", "dur_2": "15h30", "dur_0": "15h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $482 per year. It increases by 20% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "385.6", "answer_1": "578.4", "answer_3": "581.4", "answer_0": "564.4"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.273", "p1": "0.667", "p0": "0.333", "p2": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, March 20, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 361:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=873, unit_price=682, discount_pct=0, region=EU, sku_total=1555 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "59538", "v1": "1555", "v0": "595386", "v3": "596259"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = $417; priority = low.\n\nClaim to verify: \"the status is $417\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed March 25, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 01", "date_3": "April 01", "date_1": "March 30", "date_2": "March 30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = 2026-06-15; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=346, unit_price=937, discount_pct=0, region=EU, sku_total=1283 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "324548", "v1": "1283", "v2": "32420", "v0": "324202"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 2 grams of salt per serving and serves 7 people. A shopper bought a 14 gram box and a 140 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $276. A fee of $168 is charged, then a deposit of $29 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "53", "answer_1": "46", "answer_3": "50", "answer_2": "71"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 15:00 to 16:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "0h15", "dur_2": "1h00", "dur_0": "0h45", "dur_1": "1h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = $23; priority = high.\n\nClaim to verify: \"the date is $23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $893. A fee of $53 is charged, then a deposit of $71 arrives. A second fee of $26 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "885", "answer_3": "863", "answer_2": "871", "answer_1": "876"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $814. A fee of $130 is charged, then a deposit of $82 arrives. A second fee of $65 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "727", "answer_3": "676", "answer_0": "701", "answer_2": "731"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = $726; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = ops; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = $298; priority = low.\n\nClaim to verify: \"the date is $298\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.250", "p3": "0.083", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: status = 2026-03-24; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed February 12, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 18", "date_0": "February 18", "date_1": "February 16", "date_2": "February 16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $930 per year. It increases by 15% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "790.5", "answer_0": "1036.5", "answer_1": "1069.5", "answer_3": "1063.5"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=876, unit_price=721, discount_pct=0, region=EU, sku_total=1597 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "632472", "v2": "63159", "v0": "631596", "v1": "1597"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed August 05, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "August 11", "date_2": "August 13", "date_3": "August 11", "date_1": "August 09"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=694, unit_price=161, discount_pct=0, region=EU, sku_total=855 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "111040", "v2": "11173", "v1": "855", "v0": "111734"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = $717; priority = low.\n\nClaim to verify: \"the date is $717\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1400 per year. It increases by 10% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1260", "answer_1": "1540", "answer_0": "1524", "answer_3": "1504"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, July 25, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.714", "p1": "0.714", "p0": "0.286", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 216:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $20k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = $635; priority = high.\n\nClaim to verify: \"the owner is $635\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $15k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p1": "0.500", "p3": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = $162; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, February 27, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $560. A fee of $155 is charged, then a deposit of $81 arrives. A second fee of $77 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "435", "answer_2": "406", "answer_0": "409", "answer_1": "420"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $268. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=412, unit_price=959, discount_pct=0, region=EU, sku_total=1371 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "39510", "v3": "394696", "v1": "1371", "v0": "395108"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 7 people. A shopper bought a 49 gram box and a 490 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1988. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "37", "age_3": "35", "age_1": "36", "age_2": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 925:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = 2026-03-09; priority = high.\n\nClaim to verify: \"the amount is 2026-03-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed April 21, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 04", "date_1": "April 30", "date_3": "May 02", "date_2": "May 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.167", "p0": "0.250", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 820:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $267 per year. It increases by 25% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "303.75", "answer_1": "333.75", "answer_3": "317.75", "answer_2": "200.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $31k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = 2026-07-08; priority = low.\n\nClaim to verify: \"the status is 2026-07-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=181, unit_price=782, discount_pct=0, region=EU, sku_total=963 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "141542", "v2": "14154", "v3": "141723", "v1": "963"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed February 06, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 18", "date_1": "February 12", "date_0": "February 16", "date_3": "February 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 09:15 to 14:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "4h15", "dur_3": "4h00", "dur_1": "5h00", "dur_2": "4h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 8 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $739 per year. It increases by 5% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "775.95", "answer_3": "751.95", "answer_0": "760.95", "answer_2": "702.05"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 5 grams of salt per serving and serves 12 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 11 people. A shopper bought a 99 gram box and a 990 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, September 23, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1971. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "53", "age_2": "54", "age_1": "52", "age_3": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 4 grams of salt per serving and serves 5 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=858, unit_price=412, discount_pct=0, region=EU, sku_total=1270 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1270", "v0": "353496", "v3": "352638", "v2": "35349"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Wednesday, September 23, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1443 per year. It increases by 15% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1620.45", "answer_3": "1629.45", "answer_1": "1659.45", "answer_2": "1226.55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: status = 2026-02-19; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 353 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "252", "answer_1": "245", "answer_2": "260", "answer_3": "262"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1193 per year. It increases by 25% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1491.25", "answer_2": "894.75", "answer_0": "1450.25", "answer_3": "1490.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 541:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $26k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 6 grams of salt per serving and serves 11 people. A shopper bought a 66 gram box and a 660 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1961. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "62", "age_3": "61", "age_2": "64", "age_0": "63"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $463. A fee of $99 is charged, then a deposit of $37 arrives. A second fee of $49 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "352", "answer_1": "334", "answer_3": "347", "answer_2": "357"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $773. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed June 19, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 01", "date_1": "June 25", "date_3": "June 27", "date_0": "June 29"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "25", "answer_0": "45", "answer_2": "20", "answer_1": "25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $880. A fee of $109 is charged, then a deposit of $70 arrives. A second fee of $54 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "812", "answer_1": "780", "answer_0": "787", "answer_2": "806"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = 2026-04-19; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1982. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "43", "age_3": "44", "age_1": "41", "age_0": "42"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1921 per year. It increases by 20% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1536.8", "answer_0": "2280.2", "answer_1": "2305.2", "answer_3": "2307.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $960 per year. It increases by 25% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "720", "answer_3": "1175", "answer_1": "1200", "answer_0": "1157"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 6 grams of salt per serving and serves 10 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 7 people. A shopper bought a 14 gram box and a 140 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed August 09, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "August 15", "date_3": "August 17", "date_1": "August 15", "date_0": "August 17"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = pending; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = $799; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.050", "p2": "0.150", "p0": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1771 per year. It increases by 10% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1948.1", "answer_3": "1964.1", "answer_2": "1593.9", "answer_0": "1933.1"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: amount = $788; priority = high.\n\nClaim to verify: \"the amount is $788\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $839. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=963, unit_price=639, discount_pct=0, region=EU, sku_total=1602 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1602", "v2": "61535", "v0": "615357", "v3": "614394"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p1": "0.500", "p3": "0.444"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1549 per year. It increases by 10% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1394.1", "answer_1": "1703.9", "answer_0": "1685.9", "answer_3": "1690.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 310:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.200", "p2": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = $962; priority = high.\n\nClaim to verify: \"the date is $962\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_1": "30", "answer_3": "48", "answer_2": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.538", "p0": "0.462", "p2": "0.538", "p3": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 658:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = 2026-09-06; priority = high.\n\nClaim to verify: \"the amount is 2026-09-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.167", "p0": "0.333", "p1": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, January 14, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 66 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "54", "answer_1": "65", "answer_2": "64", "answer_3": "44"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $882. A fee of $99 is charged, then a deposit of $27 arrives. A second fee of $49 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "759", "answer_1": "738", "answer_3": "776", "answer_0": "761"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=320, unit_price=794, discount_pct=0, region=EU, sku_total=1114 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1114", "v2": "25408", "v3": "254400", "v0": "254080"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = 2026-10-22; priority = low.\n\nClaim to verify: \"the owner is 2026-10-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = $226; priority = low.\n\nClaim to verify: \"the owner is $226\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p2": "0.500", "p3": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = $789; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 193 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "240", "answer_2": "247", "answer_3": "222", "answer_1": "223"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.444", "p3": "0.500", "p1": "0.444", "p0": "0.556"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $321 per year. It increases by 25% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "389.25", "answer_3": "359.25", "answer_1": "401.25", "answer_2": "240.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.417", "p3": "0.545", "p0": "0.583", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 521:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 12 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=508, unit_price=966, discount_pct=0, region=EU, sku_total=1474 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1474", "v0": "490728", "v3": "491236", "v2": "49072"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = $30; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "16", "answer_2": "32", "answer_1": "34", "answer_0": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "10", "answer_1": "17", "answer_2": "15", "answer_0": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1968. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "56", "age_1": "55", "age_2": "57", "age_3": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $803. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 336:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.538", "p3": "0.417", "p1": "0.538", "p0": "0.462"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=307, unit_price=387, discount_pct=0, region=EU, sku_total=694 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "694", "v3": "119116", "v2": "11880", "v0": "118809"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = 2026-09-03; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed January 18, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 28", "date_2": "January 26", "date_1": "January 26", "date_0": "January 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1964. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "62", "age_2": "63", "age_1": "61", "age_3": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.083", "p2": "0.167", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p1": "0.667", "p2": "0.667", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $443 per year. It increases by 5% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "465.15", "answer_3": "455.15", "answer_2": "420.85", "answer_0": "419.15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1980. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "48", "age_2": "47", "age_1": "45", "age_0": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = 2026-01-14; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed May 28, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 10", "date_1": "June 04", "date_3": "June 06", "date_0": "June 08"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=373, unit_price=138, discount_pct=0, region=EU, sku_total=511 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "511", "v0": "51474", "v2": "5147", "v3": "51847"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p2": "0.667", "p0": "0.333", "p1": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=207, unit_price=687, discount_pct=0, region=EU, sku_total=894 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "14220", "v1": "894", "v3": "142416", "v0": "142209"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 09:00 to 13:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "4h45", "dur_1": "4h45", "dur_0": "4h15", "dur_3": "3h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = 2026-03-07; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 205 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "271", "answer_1": "292", "answer_3": "297", "answer_0": "280"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = $912; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 585:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "16", "answer_2": "12", "answer_0": "24", "answer_3": "12"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $650. A fee of $103 is charged, then a deposit of $70 arrives. A second fee of $51 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "566", "answer_2": "580", "answer_1": "577", "answer_3": "584"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 64 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "128", "answer_2": "146", "answer_3": "113", "answer_1": "113"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.050", "p0": "0.100", "p1": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $20k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = $480; priority = low.\n\nClaim to verify: \"the status is $480\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=160, unit_price=170, discount_pct=0, region=EU, sku_total=330 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "330", "v0": "27200", "v2": "2720", "v3": "27040"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $589 per year. It increases by 15% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "677.35", "answer_0": "647.35", "answer_3": "608.35", "answer_2": "500.65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed May 03, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 13", "date_0": "May 11", "date_3": "May 11", "date_1": "May 09"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed June 12, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 24", "date_1": "June 22", "date_0": "June 26", "date_2": "June 28"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, June 07, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.600", "p3": "0.500", "p1": "0.400", "p2": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1964. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "60", "age_3": "62", "age_1": "59", "age_2": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.778", "p3": "0.125", "p0": "0.222", "p1": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed September 19, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 01", "date_2": "September 30", "date_0": "October 02", "date_1": "September 29"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: status = 2026-06-11; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1958. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "67", "age_3": "70", "age_2": "69", "age_0": "68"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1989. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "37", "age_1": "36", "age_3": "35", "age_2": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, August 08, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p0": "0.100", "p3": "0.050", "p2": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1956 per year. It increases by 5% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2057.8", "answer_0": "2031.8", "answer_2": "1858.2", "answer_1": "2053.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, July 20, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 780:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 8 grams of salt per serving and serves 10 people. A shopper bought a 80 gram box and a 800 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = 2026-06-09; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 271 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "104", "answer_0": "96", "answer_2": "110", "answer_1": "112"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 12 people. A shopper bought a 108 gram box and a 1080 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 9 grams of salt per serving and serves 7 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = closed; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.385", "p3": "0.333", "p1": "0.615", "p2": "0.615"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.444", "p3": "0.500", "p1": "0.444", "p0": "0.556"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 11 people. A shopper bought a 44 gram box and a 440 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, May 24, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = $77; priority = high.\n\nClaim to verify: \"the status is $77\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = ops; priority = low.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 3 grams of salt per serving and serves 12 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 476:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $4k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 4 grams of salt per serving and serves 9 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1661 per year. It increases by 5% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1689.05", "answer_0": "1727.05", "answer_1": "1744.05", "answer_2": "1577.95"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $565. A fee of $128 is charged, then a deposit of $71 arrives. A second fee of $64 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "466", "answer_3": "427", "answer_0": "444", "answer_1": "467"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1986. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "37", "age_3": "36", "age_2": "39", "age_0": "38"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1960. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "65", "age_0": "66", "age_3": "64", "age_2": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p2": "0.250", "p3": "0.083", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $259. A fee of $89 is charged, then a deposit of $57 arrives. A second fee of $44 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "183", "answer_1": "191", "answer_3": "166", "answer_2": "199"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 282 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "28", "answer_3": "41", "answer_2": "42", "answer_1": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p1": "0.150", "p2": "0.050", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $208 per year. It increases by 25% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "249", "answer_2": "156", "answer_1": "260", "answer_3": "227"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 154 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "109", "answer_2": "95", "answer_1": "95", "answer_0": "90"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1965. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "61", "age_0": "60", "age_1": "59", "age_3": "62"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 970:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $623. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=812, unit_price=837, discount_pct=0, region=EU, sku_total=1649 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1649", "v0": "679644", "v2": "67964", "v3": "680456"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1873 per year. It increases by 25% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2341.25", "answer_0": "2322.25", "answer_2": "1404.75", "answer_3": "2348.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 5 people. A shopper bought a 25 gram box and a 250 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=874, unit_price=387, discount_pct=0, region=EU, sku_total=1261 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "337364", "v0": "338238", "v2": "33823", "v1": "1261"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 224 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "104", "answer_0": "108", "answer_2": "105", "answer_3": "96"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = 2026-05-05; priority = high.\n\nClaim to verify: \"the owner is 2026-05-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p3": "0.083", "p0": "0.167", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1152 per year. It increases by 5% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1094.4", "answer_3": "1184.6", "answer_1": "1209.6", "answer_0": "1197.6"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $202. A fee of $197 is charged, then a deposit of $41 arrives. A second fee of $98 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "-23", "answer_1": "-54", "answer_0": "-52", "answer_2": "-41"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.333", "p0": "0.250", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed December 24, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 28", "date_3": "December 30", "date_2": "December 28", "date_0": "December 30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1970. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "56", "age_0": "55", "age_3": "53", "age_1": "54"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed April 12, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 18", "date_0": "April 20", "date_3": "April 20", "date_2": "April 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed August 25, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 02", "date_3": "August 31", "date_0": "August 31", "date_1": "August 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $751 per year. It increases by 15% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "845.65", "answer_3": "820.65", "answer_1": "863.65", "answer_2": "638.35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed January 18, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 30", "date_1": "January 26", "date_0": "January 28", "date_3": "January 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=541, unit_price=193, discount_pct=0, region=EU, sku_total=734 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "104413", "v2": "10441", "v3": "103872", "v1": "734"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 235:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $847. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p3": "0.125", "p0": "0.250", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p0": "0.150", "p3": "0.050", "p1": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A shift runs from 12:15 to 17:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "4h15", "dur_2": "5h15", "dur_1": "5h15", "dur_0": "5h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.167", "p2": "0.333", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = 2026-05-17; priority = high.\n\nClaim to verify: \"the owner is 2026-05-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $243 per year. It increases by 15% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "279.45", "answer_3": "253.45", "answer_2": "206.55", "answer_0": "235.45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 133 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "83", "answer_0": "98", "answer_1": "95", "answer_3": "88"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, June 21, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1981. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "42", "age_2": "45", "age_1": "43", "age_0": "44"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = 2026-06-23; priority = low.\n\nClaim to verify: \"the date is 2026-06-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_3": "15", "answer_1": "17", "answer_0": "27"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = 2026-12-10; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 349 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "309", "answer_0": "320", "answer_1": "335", "answer_2": "303"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.600", "p0": "0.400", "p1": "0.600", "p3": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p3": "0.125", "p0": "0.250", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1478 per year. It increases by 5% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1494.9", "answer_2": "1404.1", "answer_1": "1551.9", "answer_0": "1533.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = 2026-04-06; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=297, unit_price=392, discount_pct=0, region=EU, sku_total=689 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "11642", "v0": "116424", "v3": "116721", "v1": "689"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_0": "48", "answer_3": "36", "answer_1": "18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.429", "p2": "0.429", "p3": "0.538", "p0": "0.571"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p1": "0.625", "p0": "0.500", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p1": "0.500", "p3": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 746:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1959. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "64", "age_0": "66", "age_2": "67", "age_1": "65"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.125", "p0": "0.375", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A shift runs from 08:15 to 12:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "3h30", "dur_0": "4h15", "dur_2": "4h45", "dur_1": "4h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = $967; priority = high.\n\nClaim to verify: \"the status is $967\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=816, unit_price=832, discount_pct=0, region=EU, sku_total=1648 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1648", "v3": "679728", "v2": "67891", "v0": "678912"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $462. A fee of $33 is charged, then a deposit of $26 arrives. A second fee of $16 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "439", "answer_3": "442", "answer_2": "466", "answer_1": "465"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=571, unit_price=825, discount_pct=0, region=EU, sku_total=1396 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1396", "v0": "471075", "v3": "471646", "v2": "47107"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p3": "0.050", "p1": "0.200", "p2": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=799, unit_price=984, discount_pct=0, region=EU, sku_total=1783 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "785417", "v2": "78621", "v1": "1783", "v0": "786216"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Thursday, July 02, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 378 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "60", "answer_0": "63", "answer_2": "53", "answer_1": "60"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, November 02, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = 2026-03-05; priority = low.\n\nClaim to verify: \"the amount is 2026-03-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = 2026-05-17; priority = low.\n\nClaim to verify: \"the status is 2026-05-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed February 19, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 01", "date_1": "February 27", "date_3": "March 01", "date_0": "March 03"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=560, unit_price=545, discount_pct=0, region=EU, sku_total=1105 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1105", "v0": "305200", "v3": "304640", "v2": "30520"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1979. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "44", "age_0": "46", "age_1": "45", "age_2": "47"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 942:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.500", "p3": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $471. A fee of $173 is charged, then a deposit of $23 arrives. A second fee of $86 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "235", "answer_1": "230", "answer_3": "244", "answer_2": "207"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = 2026-10-05; priority = high.\n\nClaim to verify: \"the owner is 2026-10-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 07:30 to 18:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "10h15", "dur_0": "11h00", "dur_2": "11h30", "dur_1": "11h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 87 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "57", "answer_0": "54", "answer_3": "57", "answer_2": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = $111; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, August 15, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p1": "0.500", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = $659; priority = low.\n\nClaim to verify: \"the status is $659\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $523. A fee of $180 is charged, then a deposit of $72 arrives. A second fee of $90 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "333", "answer_0": "325", "answer_1": "321", "answer_2": "354"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 665:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p1": "0.625", "p0": "0.500", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed October 15, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 24", "date_1": "October 20", "date_0": "October 22", "date_3": "October 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 2 grams of salt per serving and serves 12 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.727", "p0": "0.273", "p3": "0.200", "p1": "0.727"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 288:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.083", "p2": "0.167", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 828:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.286", "p0": "0.375", "p2": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=389, unit_price=847, discount_pct=0, region=EU, sku_total=1236 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "329483", "v3": "329094", "v1": "1236", "v2": "32948"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = 2026-08-08; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 283 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "84", "answer_2": "94", "answer_0": "96", "answer_1": "113"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p0": "0.250", "p3": "0.125", "p1": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=567, unit_price=996, discount_pct=0, region=EU, sku_total=1563 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1563", "v0": "564732", "v3": "564165", "v2": "56473"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = $38; priority = high.\n\nClaim to verify: \"the amount is $38\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $768. A fee of $138 is charged, then a deposit of $44 arrives. A second fee of $69 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "605", "answer_1": "599", "answer_3": "608", "answer_2": "616"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed December 19, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 30", "date_3": "December 31", "date_0": "January 01", "date_1": "December 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed August 26, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 31", "date_1": "August 29", "date_2": "August 29", "date_0": "August 31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=636, unit_price=681, discount_pct=0, region=EU, sku_total=1317 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "432480", "v1": "1317", "v2": "43311", "v0": "433116"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = pending; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 749:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $3k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_1": "30", "answer_2": "24", "answer_3": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.167", "p1": "0.500", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 259 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "144", "answer_1": "134", "answer_3": "161", "answer_2": "140"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 172 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "276", "answer_2": "262", "answer_0": "280", "answer_1": "288"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = closed; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "32", "answer_2": "25", "answer_3": "35", "answer_0": "60"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 658:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = 2026-08-19; priority = high.\n\nClaim to verify: \"the owner is 2026-08-19\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, August 24, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=765, unit_price=652, discount_pct=0, region=EU, sku_total=1417 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "498780", "v2": "49878", "v3": "499545", "v1": "1417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 12 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = pending; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $729. A fee of $67 is charged, then a deposit of $84 arrives. A second fee of $33 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "730", "answer_0": "713", "answer_2": "734", "answer_1": "702"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1260 per year. It increases by 10% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1350", "answer_1": "1386", "answer_2": "1134", "answer_3": "1324"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $12k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 2 grams of salt per serving and serves 9 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 495:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = 2026-12-07; priority = high.\n\nClaim to verify: \"the date is 2026-12-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: owner = 2026-08-26; priority = low.\n\nClaim to verify: \"the owner is 2026-08-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Thursday, April 16, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1966. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_1": "57", "age_2": "59", "age_3": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $352. A fee of $146 is charged, then a deposit of $65 arrives. A second fee of $73 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "217", "answer_1": "212", "answer_0": "198", "answer_2": "187"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1033 per year. It increases by 10% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1106.3", "answer_2": "929.7", "answer_3": "1076.3", "answer_1": "1136.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1556 per year. It increases by 5% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1478.2", "answer_0": "1589.8", "answer_1": "1633.8", "answer_3": "1555.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 985:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 4 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 924:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A shift runs from 06:00 to 21:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "14h30", "dur_2": "15h15", "dur_1": "15h30", "dur_0": "15h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = $781; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = $790; priority = high.\n\nClaim to verify: \"the amount is $790\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p3": "0.385", "p2": "0.571", "p0": "0.429"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1981. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "45", "age_1": "43", "age_0": "44", "age_3": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $455 per year. It increases by 5% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "477.75", "answer_2": "432.25", "answer_3": "453.75", "answer_0": "460.75"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A shift runs from 14:30 to 20:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "6h15", "dur_1": "6h15", "dur_3": "5h15", "dur_0": "5h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=208, unit_price=572, discount_pct=0, region=EU, sku_total=780 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "118976", "v2": "11897", "v1": "780", "v3": "119184"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1958. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "66", "age_3": "65", "age_2": "68", "age_0": "67"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.800", "p0": "0.200", "p3": "0.111", "p1": "0.800"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed December 23, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 03", "date_3": "January 03", "date_1": "January 01", "date_0": "January 05"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 363:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 896:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 878:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1988. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "36", "age_0": "37", "age_2": "38", "age_3": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_0": "42", "answer_1": "22", "answer_3": "24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 16:45 to 19:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "2h30", "dur_1": "2h30", "dur_3": "1h30", "dur_0": "2h00"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=561, unit_price=593, discount_pct=0, region=EU, sku_total=1154 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "332673", "v2": "33267", "v1": "1154", "v3": "333234"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=122, unit_price=532, discount_pct=0, region=EU, sku_total=654 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "64904", "v1": "654", "v2": "6490", "v3": "65026"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 9 grams of salt per serving and serves 4 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 774:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 365:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 10:00 to 18:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "8h45", "dur_3": "7h30", "dur_1": "8h30", "dur_0": "8h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1963. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "63", "age_2": "64", "age_1": "62", "age_3": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 12:00 to 18:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "6h45", "dur_0": "6h30", "dur_2": "7h00", "dur_3": "5h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 183 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "69", "answer_3": "81", "answer_2": "75", "answer_0": "72"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 807:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, July 25, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=484, unit_price=989, discount_pct=0, region=EU, sku_total=1473 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "478192", "v1": "1473", "v0": "478676", "v2": "47867"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 5 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=964, unit_price=582, discount_pct=0, region=EU, sku_total=1546 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1546", "v0": "561048", "v2": "56104", "v3": "562012"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, June 25, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Friday, June 19, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $28k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $225. A fee of $95 is charged, then a deposit of $49 arrives. A second fee of $47 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "132", "answer_2": "115", "answer_3": "123", "answer_1": "115"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1112 per year. It increases by 25% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "834", "answer_3": "1308", "answer_1": "1390", "answer_0": "1343"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1970. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "54", "age_1": "55", "age_0": "56", "age_2": "57"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.333", "p1": "0.417", "p3": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p0": "0.250", "p3": "0.125", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed January 15, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 24", "date_0": "January 26", "date_1": "January 22", "date_3": "January 24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 09:00 to 11:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h00", "dur_3": "1h45", "dur_2": "2h15", "dur_1": "2h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.462", "p3": "0.500", "p0": "0.538", "p1": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $730. A fee of $150 is charged, then a deposit of $52 arrives. A second fee of $75 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "557", "answer_3": "560", "answer_2": "576", "answer_1": "547"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = $690; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_0": "50", "answer_1": "22", "answer_2": "15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, September 20, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 176 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "132", "answer_3": "137", "answer_1": "156", "answer_0": "150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed May 03, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 12", "date_0": "May 14", "date_2": "May 16", "date_3": "May 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $753. A fee of $121 is charged, then a deposit of $66 arrives. A second fee of $60 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "609", "answer_0": "638", "answer_1": "609", "answer_2": "614"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = $286; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "53", "age_0": "52", "age_1": "51", "age_3": "54"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 8 grams of salt per serving and serves 6 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = $523; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed March 17, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 20", "date_3": "March 22", "date_1": "March 20", "date_2": "March 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.400", "p0": "0.600", "p3": "0.500", "p1": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 11:45 to 14:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "3h00", "dur_2": "3h00", "dur_0": "2h30", "dur_3": "2h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=974, unit_price=870, discount_pct=0, region=EU, sku_total=1844 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "84738", "v0": "847380", "v3": "848354", "v1": "1844"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1967. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "60", "age_3": "57", "age_1": "58", "age_0": "59"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 328:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 381 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "110", "answer_0": "120", "answer_3": "131", "answer_1": "134"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 228:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Sunday, October 25, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = 2026-01-13; priority = high.\n\nClaim to verify: \"the amount is 2026-01-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_0": "36", "answer_1": "26", "answer_2": "24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 90 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "84", "answer_3": "91", "answer_1": "65", "answer_2": "81"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.222", "p3": "0.125", "p2": "0.778", "p1": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed April 13, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 27", "date_3": "April 25", "date_2": "April 25", "date_1": "April 23"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1235 per year. It increases by 5% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1271.75", "answer_3": "1296.75", "answer_2": "1173.25", "answer_1": "1296.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 7 grams of salt per serving and serves 6 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $211. A fee of $194 is charged, then a deposit of $80 arrives. A second fee of $97 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "0", "answer_2": "21", "answer_3": "3", "answer_1": "-12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=773, unit_price=469, discount_pct=0, region=EU, sku_total=1242 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "362537", "v3": "363310", "v1": "1242", "v2": "36253"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 830:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = 2026-06-26; priority = low.\n\nClaim to verify: \"the amount is 2026-06-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $751. A fee of $199 is charged, then a deposit of $48 arrives. A second fee of $99 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "489", "answer_0": "501", "answer_2": "526", "answer_3": "476"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 140 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "37", "answer_2": "31", "answer_0": "45", "answer_3": "29"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $409. A fee of $159 is charged, then a deposit of $82 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "258", "answer_1": "234", "answer_2": "274", "answer_0": "253"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = 2026-02-02; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = $263; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 292 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "74", "answer_2": "54", "answer_1": "64", "answer_0": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = open; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=487, unit_price=427, discount_pct=0, region=EU, sku_total=914 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "914", "v3": "207462", "v0": "207949", "v2": "20794"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.167", "p2": "0.167", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 184 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "124", "answer_2": "108", "answer_1": "146", "answer_0": "128"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p0": "0.100", "p3": "0.050", "p1": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1958. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "66", "age_1": "65", "age_3": "64", "age_2": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $316. A fee of $103 is charged, then a deposit of $50 arrives. A second fee of $51 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "236", "answer_1": "225", "answer_0": "212", "answer_3": "237"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = $539; priority = low.\n\nClaim to verify: \"the owner is $539\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Sunday, December 20, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed April 01, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 12", "date_2": "April 12", "date_0": "April 14", "date_1": "April 10"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.375", "p1": "0.500", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, August 17, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $419 per year. It increases by 10% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "377.1", "answer_0": "438.9", "answer_3": "472.9", "answer_1": "460.9"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1960. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "66", "age_0": "65", "age_1": "64", "age_3": "63"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed December 22, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 29", "date_2": "January 02", "date_0": "December 31", "date_3": "December 31"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.375", "p3": "0.125", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=120, unit_price=761, discount_pct=0, region=EU, sku_total=881 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "91200", "v1": "881", "v2": "9132", "v0": "91320"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $212. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_1": "31", "answer_0": "52", "answer_3": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 737:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1963. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "62", "age_0": "63", "age_3": "61", "age_2": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p3": "0.125", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p3": "0.050", "p2": "0.050", "p0": "0.100"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=719, unit_price=891, discount_pct=0, region=EU, sku_total=1610 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "640629", "v3": "639910", "v2": "64062", "v1": "1610"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $4k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = $806; priority = high.\n\nClaim to verify: \"the owner is $806\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 07:00 to 20:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "12h00", "dur_1": "13h00", "dur_2": "13h00", "dur_0": "12h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $464. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed June 06, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 15", "date_1": "June 14", "date_3": "June 16", "date_0": "June 17"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $13k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1147 per year. It increases by 5% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1156.35", "answer_0": "1176.35", "answer_1": "1204.35", "answer_2": "1089.65"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: date = 2026-03-16; priority = low.\n\nClaim to verify: \"the date is 2026-03-16\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $861. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 368:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed January 06, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 09", "date_0": "January 09", "date_2": "January 11", "date_3": "January 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 06:30 to 17:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "10h45", "dur_3": "10h00", "dur_0": "10h30", "dur_1": "11h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=199, unit_price=619, discount_pct=0, region=EU, sku_total=818 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "122982", "v0": "123181", "v1": "818", "v2": "12318"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 10 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 835:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=613, unit_price=498, discount_pct=0, region=EU, sku_total=1111 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "305274", "v2": "30527", "v3": "305887", "v1": "1111"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = $37; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = 2026-12-05; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, March 24, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = 2026-11-24; priority = low.\n\nClaim to verify: \"the date is 2026-11-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 360:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed May 06, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 16", "date_1": "May 14", "date_0": "May 18", "date_2": "May 16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $40k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 8 grams of salt per serving and serves 11 people. A shopper bought a 88 gram box and a 880 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed May 22, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "June 05", "date_3": "June 03", "date_1": "June 01", "date_2": "June 03"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = $850; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed December 12, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 21", "date_3": "December 20", "date_2": "December 23", "date_1": "December 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Thursday, October 01, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1985. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "41", "age_0": "40", "age_1": "39", "age_3": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = 2026-03-03; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "27", "answer_2": "15", "answer_1": "19", "answer_3": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $319 per year. It increases by 20% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "355.8", "answer_1": "382.8", "answer_0": "364.8", "answer_2": "255.2"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $837. A fee of $173 is charged, then a deposit of $38 arrives. A second fee of $86 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "596", "answer_3": "639", "answer_2": "612", "answer_0": "616"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 14:15 to 20:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "5h15", "dur_1": "6h15", "dur_0": "6h00", "dur_2": "6h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_0": "32", "answer_2": "20", "answer_1": "23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1967 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2033.35", "answer_2": "1868.65", "answer_3": "2038.35", "answer_1": "2065.35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.375", "p0": "0.250", "p2": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1983. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "41", "age_3": "39", "age_2": "42", "age_1": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.429", "p2": "0.533", "p0": "0.467", "p1": "0.533"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 682:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_2": "32", "answer_0": "56", "answer_1": "35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed March 12, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 16", "date_0": "March 18", "date_2": "March 16", "date_3": "March 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=969, unit_price=544, discount_pct=0, region=EU, sku_total=1513 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "52713", "v0": "527136", "v1": "1513", "v3": "528105"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 211 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "177", "answer_0": "180", "answer_2": "161", "answer_1": "200"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p1": "0.500", "p3": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $829 per year. It increases by 25% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "991.25", "answer_1": "1036.25", "answer_3": "1021.25", "answer_2": "621.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 8 grams of salt per serving and serves 9 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 11:00 to 19:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "8h30", "dur_2": "8h15", "dur_3": "7h30", "dur_0": "7h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.385", "p2": "0.385", "p3": "0.583", "p0": "0.615"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1133 per year. It increases by 20% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1316.6", "answer_3": "1287.6", "answer_1": "1359.6", "answer_2": "906.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed September 14, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 23", "date_2": "September 21", "date_1": "September 21", "date_0": "September 23"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = $365; priority = low.\n\nClaim to verify: \"the amount is $365\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = $65; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $281. A fee of $31 is charged, then a deposit of $59 arrives. A second fee of $15 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "275", "answer_0": "294", "answer_1": "278", "answer_3": "274"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed December 15, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 23", "date_0": "December 25", "date_3": "December 25", "date_2": "December 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed February 23, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 28", "date_2": "March 04", "date_0": "March 02", "date_3": "March 02"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1745 per year. It increases by 10% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1917.5", "answer_1": "1919.5", "answer_0": "1908.5", "answer_2": "1570.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1982. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "44", "age_0": "43", "age_3": "45", "age_1": "42"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 11:15 to 15:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "3h15", "dur_0": "3h30", "dur_2": "4h00", "dur_1": "4h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 16:00 to 17:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "1h15", "dur_2": "1h15", "dur_0": "1h00", "dur_3": "0h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, April 05, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 10 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.083", "p0": "0.167", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A shift runs from 12:15 to 16:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "3h00", "dur_0": "3h45", "dur_1": "4h00", "dur_2": "4h00"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_1": "34", "answer_2": "28", "answer_0": "70"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "49", "answer_0": "77", "answer_2": "28", "answer_1": "35"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 2 grams of salt per serving and serves 4 people. A shopper bought a 8 gram box and a 80 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1174 per year. It increases by 20% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1332.8", "answer_1": "1408.8", "answer_0": "1362.8", "answer_2": "939.2"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1966 per year. It increases by 10% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2135.6", "answer_1": "2162.6", "answer_3": "2109.6", "answer_2": "1769.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=529, unit_price=790, discount_pct=0, region=EU, sku_total=1319 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1319", "v3": "418439", "v2": "41791", "v0": "417910"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.571", "p2": "0.375", "p1": "0.375", "p0": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed July 11, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 20", "date_3": "July 19", "date_1": "July 17", "date_2": "July 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1990. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "35", "age_2": "37", "age_0": "36", "age_3": "38"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1984. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "42", "age_2": "41", "age_1": "39", "age_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = 2026-11-15; priority = low.\n\nClaim to verify: \"the status is 2026-11-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $227. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = $177; priority = low.\n\nClaim to verify: \"the date is $177\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 351 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "105", "answer_3": "103", "answer_1": "117", "answer_2": "98"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed January 27, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 06", "date_3": "February 08", "date_2": "February 12", "date_0": "February 10"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 09:00 to 12:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h30", "dur_3": "2h30", "dur_0": "3h15", "dur_1": "3h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = 2026-01-12; priority = low.\n\nClaim to verify: \"the amount is 2026-01-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: owner = 2026-09-09; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $385. A fee of $91 is charged, then a deposit of $44 arrives. A second fee of $45 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "285", "answer_0": "293", "answer_2": "303", "answer_1": "310"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_2": "8", "answer_1": "12", "answer_3": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "24", "answer_3": "20", "answer_0": "40", "answer_2": "20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, August 21, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $239. A fee of $170 is charged, then a deposit of $55 arrives. A second fee of $85 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "61", "answer_0": "39", "answer_1": "9", "answer_2": "64"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, July 22, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=580, unit_price=832, discount_pct=0, region=EU, sku_total=1412 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "482560", "v2": "48256", "v1": "1412", "v3": "481980"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 804:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $393. A fee of $128 is charged, then a deposit of $36 arrives. A second fee of $64 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "225", "answer_0": "237", "answer_3": "251", "answer_2": "264"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.667", "p2": "0.286", "p1": "0.286", "p0": "0.714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 510:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = 2026-05-20; priority = high.\n\nClaim to verify: \"the owner is 2026-05-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=172, unit_price=842, discount_pct=0, region=EU, sku_total=1014 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "144652", "v2": "14482", "v0": "144824", "v1": "1014"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $14k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $15k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.500", "p3": "0.167", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1610 per year. It increases by 25% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1207.5", "answer_3": "1981.5", "answer_0": "1992.5", "answer_1": "2012.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $5k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 12, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 26", "date_1": "April 22", "date_0": "April 24", "date_3": "April 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1963. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "61", "age_3": "64", "age_2": "63", "age_0": "62"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $827. A fee of $84 is charged, then a deposit of $32 arrives. A second fee of $42 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "737", "answer_1": "745", "answer_0": "733", "answer_2": "750"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.150", "p2": "0.050", "p0": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1752 per year. It increases by 20% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2073.4", "answer_2": "1401.6", "answer_3": "2081.4", "answer_1": "2102.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p2": "0.250", "p1": "0.500", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $762 per year. It increases by 20% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "893.4", "answer_1": "914.4", "answer_0": "869.4", "answer_2": "609.6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 477:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 377 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "40", "answer_0": "42", "answer_3": "56", "answer_2": "22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.333", "p3": "0.625", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 09:00 to 15:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "6h00", "dur_3": "5h45", "dur_1": "6h45", "dur_2": "6h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 235 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "51", "answer_1": "67", "answer_2": "30", "answer_0": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = $587; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p2": "0.125", "p0": "0.250", "p3": "0.125"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 7 grams of salt per serving and serves 10 people. A shopper bought a 70 gram box and a 700 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1173 per year. It increases by 10% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1259.3", "answer_2": "1055.7", "answer_1": "1290.3", "answer_0": "1275.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $694. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $859. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed August 17, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "August 24", "date_1": "August 24", "date_0": "August 26", "date_3": "August 26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1984. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "42", "age_1": "39", "age_0": "40", "age_2": "41"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 376:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1962. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "64", "age_1": "62", "age_0": "63", "age_3": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed August 03, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 10", "date_2": "August 08", "date_1": "August 08", "date_0": "August 10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed October 08, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "October 21", "date_2": "October 19", "date_3": "October 19", "date_1": "October 17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed May 19, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 24", "date_0": "May 26", "date_1": "May 24", "date_3": "May 26"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 390 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "65", "answer_2": "86", "answer_0": "80", "answer_1": "67"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $844. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = $777; priority = high.\n\nClaim to verify: \"the owner is $777\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 953:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.300", "p1": "0.300", "p0": "0.700", "p3": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.167", "p2": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = pending; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_2": "28", "answer_1": "32", "answer_3": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, June 14, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, April 28, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $773. A fee of $61 is charged, then a deposit of $41 arrives. A second fee of $30 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "720", "answer_3": "709", "answer_2": "696", "answer_0": "723"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_0": "66", "answer_3": "36", "answer_1": "36"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.083", "p0": "0.167", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $745 per year. It increases by 20% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "849", "answer_2": "596", "answer_1": "894", "answer_3": "836"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $450. A fee of $50 is charged, then a deposit of $38 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "413", "answer_1": "430", "answer_3": "434", "answer_2": "391"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 751:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = 2026-12-03; priority = high.\n\nClaim to verify: \"the date is 2026-12-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 98 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "128", "answer_2": "148", "answer_1": "137", "answer_3": "137"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $396. A fee of $34 is charged, then a deposit of $27 arrives. A second fee of $17 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "359", "answer_0": "372", "answer_1": "396", "answer_3": "383"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_0": "36", "answer_1": "27", "answer_3": "12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $982 per year. It increases by 15% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1129.3", "answer_2": "834.7", "answer_0": "1119.3", "answer_3": "1090.3"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 19:45 to 22:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h30", "dur_2": "3h00", "dur_3": "1h45", "dur_1": "2h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1546 per year. It increases by 15% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1314.1", "answer_3": "1745.9", "answer_1": "1777.9", "answer_0": "1740.9"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $687. A fee of $124 is charged, then a deposit of $25 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "526", "answer_1": "498", "answer_2": "534", "answer_3": "510"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 92 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "206", "answer_1": "173", "answer_0": "192", "answer_3": "186"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $406. A fee of $133 is charged, then a deposit of $73 arrives. A second fee of $66 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "283", "answer_1": "265", "answer_2": "272", "answer_0": "280"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 119 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "113", "answer_0": "96", "answer_3": "112", "answer_2": "80"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1989. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "34", "age_3": "33", "age_2": "36", "age_0": "35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.333", "p3": "0.083", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $535 per year. It increases by 20% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "428", "answer_1": "642", "answer_0": "618", "answer_3": "585"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 512:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $605 per year. It increases by 20% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "484", "answer_1": "726", "answer_0": "704", "answer_3": "737"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, March 20, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = 2026-07-18; priority = low.\n\nClaim to verify: \"the amount is 2026-07-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.714", "p3": "0.167", "p1": "0.714", "p0": "0.286"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = $454; priority = high.\n\nClaim to verify: \"the amount is $454\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1964. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "60", "age_2": "61", "age_1": "59", "age_3": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p1": "0.250", "p2": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 13:00 to 19:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "6h00", "dur_3": "5h45", "dur_2": "6h30", "dur_1": "6h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = open; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = 2026-06-18; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $884. A fee of $37 is charged, then a deposit of $65 arrives. A second fee of $18 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "894", "answer_2": "911", "answer_3": "870", "answer_1": "887"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = $102; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=749, unit_price=206, discount_pct=0, region=EU, sku_total=955 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "153545", "v2": "15429", "v0": "154294", "v1": "955"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, February 26, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 360:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = $868; priority = high.\n\nClaim to verify: \"the amount is $868\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $572. A fee of $104 is charged, then a deposit of $79 arrives. A second fee of $52 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "480", "answer_3": "473", "answer_2": "522", "answer_0": "495"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 973:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.167", "p1": "0.333", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 2 grams of salt per serving and serves 11 people. A shopper bought a 22 gram box and a 220 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = pending; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 177 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "78", "answer_1": "75", "answer_0": "60", "answer_2": "77"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=455, unit_price=239, discount_pct=0, region=EU, sku_total=694 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "694", "v2": "10874", "v3": "108290", "v0": "108745"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.714", "p2": "0.714", "p0": "0.286", "p3": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed July 23, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 29", "date_0": "July 29", "date_1": "July 27", "date_2": "July 31"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_2": "12", "answer_3": "24", "answer_1": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 302:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.455", "p2": "0.500", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = $483; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 372 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "134", "answer_3": "118", "answer_0": "126", "answer_1": "141"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $690. A fee of $162 is charged, then a deposit of $42 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "489", "answer_3": "473", "answer_2": "460", "answer_1": "507"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p0": "0.429", "p3": "0.385", "p2": "0.571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, February 27, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = 2026-08-07; priority = low.\n\nClaim to verify: \"the owner is 2026-08-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = $317; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 09:45 to 11:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "0h30", "dur_2": "0h45", "dur_3": "0h15", "dur_1": "1h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $22k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = $789; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 11 people. A shopper bought a 33 gram box and a 330 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.364", "p3": "0.600", "p1": "0.364", "p0": "0.636"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.200", "p2": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1974. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "48", "age_1": "49", "age_2": "51", "age_0": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed March 16, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 24", "date_3": "March 26", "date_2": "March 28", "date_0": "March 26"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = 2026-03-15; priority = low.\n\nClaim to verify: \"the owner is 2026-03-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 112 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "38", "answer_1": "59", "answer_3": "45", "answer_0": "54"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: status = $269; priority = high.\n\nClaim to verify: \"the status is $269\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Sunday, April 05, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = $491; priority = low.\n\nClaim to verify: \"the date is $491\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p0": "0.250", "p2": "0.125", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.083", "p2": "0.250", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1885 per year. It increases by 15% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2133.75", "answer_1": "2167.75", "answer_2": "1602.25", "answer_3": "2156.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 5 people. A shopper bought a 25 gram box and a 250 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $412 per year. It increases by 25% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "464", "answer_0": "501", "answer_2": "309", "answer_1": "515"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 266 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "25", "answer_3": "47", "answer_0": "36", "answer_1": "17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.167", "p0": "0.333", "p3": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.050", "p1": "0.150", "p0": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.700", "p2": "0.273", "p1": "0.273", "p0": "0.727"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 944:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $210 per year. It increases by 20% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "168", "answer_0": "227", "answer_1": "252", "answer_3": "266"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 456:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $908 per year. It increases by 10% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "934.8", "answer_0": "973.8", "answer_1": "998.8", "answer_2": "817.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 850:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "9", "answer_1": "15", "answer_0": "21", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $767. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 294:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $14k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $679. A fee of $120 is charged, then a deposit of $55 arrives. A second fee of $60 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "554", "answer_3": "576", "answer_1": "541", "answer_2": "540"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1977. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "48", "age_3": "45", "age_0": "47", "age_1": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed March 04, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 10", "date_2": "March 08", "date_1": "March 08", "date_3": "March 10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.375", "p1": "0.625", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p2": "0.250", "p3": "0.125", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "44", "answer_3": "32", "answer_0": "72", "answer_2": "40"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.286", "p1": "0.286", "p3": "0.667", "p0": "0.714"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = $377; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 3 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1986. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "42", "age_2": "41", "age_1": "39", "age_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed December 08, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 11", "date_0": "December 11", "date_2": "December 09", "date_3": "December 13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=897, unit_price=803, discount_pct=0, region=EU, sku_total=1700 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1700", "v2": "72029", "v0": "720291", "v3": "719394"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_3": "12", "answer_0": "24", "answer_1": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1962. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "61", "age_0": "62", "age_2": "63", "age_3": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Monday, September 28, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $119. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1958. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "66", "age_2": "69", "age_0": "68", "age_1": "67"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p2": "0.083", "p1": "0.250", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $230. A fee of $58 is charged, then a deposit of $34 arrives. A second fee of $29 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "194", "answer_1": "156", "answer_2": "153", "answer_0": "177"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = 2026-08-11; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1975. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "51", "age_3": "52", "age_0": "50", "age_1": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 332 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "160", "answer_0": "150", "answer_2": "168", "answer_3": "153"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Wednesday, May 27, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1682 per year. It increases by 10% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1841.2", "answer_2": "1513.8", "answer_0": "1834.2", "answer_1": "1850.2"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 263 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "324", "answer_1": "311", "answer_2": "338", "answer_3": "335"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = $51; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 280 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "138", "answer_0": "144", "answer_2": "164", "answer_1": "152"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $865. A fee of $132 is charged, then a deposit of $83 arrives. A second fee of $66 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "730", "answer_0": "750", "answer_3": "759", "answer_1": "735"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $417 per year. It increases by 10% this year, then a $49 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "422.7", "answer_2": "375.3", "answer_0": "409.7", "answer_1": "458.7"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, October 23, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1183 per year. It increases by 15% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1362.45", "answer_0": "1324.45", "answer_1": "1360.45", "answer_2": "1005.55"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.167", "p1": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.167", "p2": "0.333", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 132 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "450", "answer_1": "415", "answer_2": "415", "answer_0": "432"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 61 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "147", "answer_2": "156", "answer_0": "150", "answer_3": "155"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.300", "p3": "0.222", "p2": "0.700", "p1": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $763. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=869, unit_price=790, discount_pct=0, region=EU, sku_total=1659 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "685641", "v0": "686510", "v2": "68651", "v1": "1659"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = $852; priority = low.\n\nClaim to verify: \"the date is $852\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p3": "0.600", "p1": "0.333", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 15:15 to 19:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "3h45", "dur_2": "4h15", "dur_1": "4h30", "dur_3": "3h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 340 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "35", "answer_0": "45", "answer_3": "36", "answer_2": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 7 grams of salt per serving and serves 7 people. A shopper bought a 49 gram box and a 490 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "15", "answer_3": "25", "answer_1": "20", "answer_0": "40"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $339. A fee of $127 is charged, then a deposit of $39 arrives. A second fee of $63 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "188", "answer_2": "206", "answer_1": "182", "answer_3": "192"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=965, unit_price=777, discount_pct=0, region=EU, sku_total=1742 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "749805", "v3": "748840", "v1": "1742", "v2": "74980"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 341 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "189", "answer_3": "162", "answer_0": "175", "answer_2": "157"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p0": "0.333", "p3": "0.083", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $898. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = $630; priority = low.\n\nClaim to verify: \"the status is $630\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $304. A fee of $53 is charged, then a deposit of $56 arrives. A second fee of $26 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "311", "answer_2": "259", "answer_0": "281", "answer_1": "275"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = 2026-07-13; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 3 grams of salt per serving and serves 12 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "9", "answer_1": "9", "answer_0": "15", "answer_2": "6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed May 24, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 30", "date_2": "May 26", "date_1": "May 28", "date_0": "May 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, April 08, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $146. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 528:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Thursday, February 12, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.714", "p1": "0.286", "p2": "0.286", "p3": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=844, unit_price=746, discount_pct=0, region=EU, sku_total=1590 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1590", "v0": "629624", "v3": "628780", "v2": "62962"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $14k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = $745; priority = low.\n\nClaim to verify: \"the amount is $745\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=754, unit_price=710, discount_pct=0, region=EU, sku_total=1464 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "536094", "v2": "53534", "v0": "535340", "v1": "1464"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = closed; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1417 per year. It increases by 20% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1677.4", "answer_1": "1700.4", "answer_3": "1668.4", "answer_2": "1133.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = $607; priority = low.\n\nClaim to verify: \"the date is $607\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 886:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = $606; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.556", "p0": "0.444", "p1": "0.556", "p3": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.800", "p3": "0.778", "p2": "0.200", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 215 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "135", "answer_2": "139", "answer_1": "130", "answer_3": "137"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.500", "p2": "0.167", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = $122; priority = high.\n\nClaim to verify: \"the status is $122\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p1": "0.200", "p0": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1932 per year. It increases by 10% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1738.8", "answer_0": "2075.2", "answer_1": "2125.2", "answer_3": "2069.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: amount = $369; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 698:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_3": "12", "answer_1": "20", "answer_0": "30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=422, unit_price=775, discount_pct=0, region=EU, sku_total=1197 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1197", "v3": "327472", "v0": "327050", "v2": "32705"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, June 16, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Sunday, April 12, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 18:15 to 21:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h00", "dur_0": "2h30", "dur_3": "2h15", "dur_1": "3h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = 2026-12-04; priority = high.\n\nClaim to verify: \"the owner is 2026-12-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $812 per year. It increases by 15% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "906.8", "answer_3": "946.8", "answer_1": "933.8", "answer_2": "690.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 506:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed August 07, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 17", "date_2": "August 19", "date_3": "August 19", "date_0": "August 21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=548, unit_price=787, discount_pct=0, region=EU, sku_total=1335 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "431824", "v2": "43127", "v0": "431276", "v1": "1335"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 175 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "54", "answer_3": "56", "answer_1": "52", "answer_2": "74"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed August 05, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 12", "date_2": "August 16", "date_0": "August 14", "date_3": "August 14"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "25", "answer_2": "10", "answer_1": "15", "answer_0": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1966. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "59", "age_1": "57", "age_0": "58", "age_3": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $654. A fee of $93 is charged, then a deposit of $58 arrives. A second fee of $46 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "573", "answer_3": "550", "answer_2": "583", "answer_1": "586"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1371 per year. It increases by 20% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1622.2", "answer_1": "1645.2", "answer_2": "1096.8", "answer_0": "1635.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.667", "p3": "0.625", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = ops; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1962. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "62", "age_1": "61", "age_2": "63", "age_3": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.667", "p0": "0.333", "p3": "0.273"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = open; priority = high.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $493. A fee of $32 is charged, then a deposit of $56 arrives. A second fee of $16 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "501", "answer_3": "488", "answer_1": "520", "answer_2": "511"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=845, unit_price=186, discount_pct=0, region=EU, sku_total=1031 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "157170", "v3": "156325", "v2": "15717", "v1": "1031"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=775, unit_price=934, discount_pct=0, region=EU, sku_total=1709 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "723075", "v1": "1709", "v0": "723850", "v2": "72385"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_1": "28", "answer_3": "32", "answer_0": "56"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed July 28, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 05", "date_0": "August 07", "date_2": "August 09", "date_3": "August 07"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $201. A fee of $41 is charged, then a deposit of $56 arrives. A second fee of $20 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "216", "answer_0": "196", "answer_2": "193", "answer_1": "204"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "14", "answer_3": "15", "answer_2": "9", "answer_0": "24"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 358:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 11 people. A shopper bought a 44 gram box and a 440 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.750", "p3": "0.143", "p0": "0.250", "p1": "0.750"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $836 per year. It increases by 20% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1003.2", "answer_0": "983.2", "answer_2": "668.8", "answer_3": "988.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = 2026-07-02; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=461, unit_price=374, discount_pct=0, region=EU, sku_total=835 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "172875", "v0": "172414", "v1": "835", "v2": "17241"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed September 23, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 01", "date_0": "October 01", "date_2": "October 03", "date_1": "September 29"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.333", "p3": "0.083", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $30k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 565:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p2": "0.050", "p0": "0.100", "p3": "0.050"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $370 per year. It increases by 10% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "333", "answer_0": "365", "answer_3": "371", "answer_1": "407"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.444", "p1": "0.556", "p2": "0.556", "p3": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 805:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p0": "0.500", "p3": "0.444"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = 2026-03-24; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed December 01, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 11", "date_0": "December 11", "date_1": "December 09", "date_2": "December 13"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $646. A fee of $134 is charged, then a deposit of $33 arrives. A second fee of $67 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "448", "answer_3": "483", "answer_2": "488", "answer_0": "478"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $972 per year. It increases by 15% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1117.8", "answer_3": "1120.8", "answer_2": "826.2", "answer_0": "1104.8"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.700", "p2": "0.300", "p3": "0.667", "p1": "0.300"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, October 17, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $392. A fee of $66 is charged, then a deposit of $23 arrives. A second fee of $33 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "300", "answer_1": "298", "answer_0": "316", "answer_2": "320"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 10 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed June 19, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 26", "date_1": "June 24", "date_2": "June 28", "date_0": "June 26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.167", "p1": "0.833", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 323 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "240", "answer_2": "246", "answer_1": "230", "answer_3": "253"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.583", "p3": "0.364", "p1": "0.583", "p0": "0.417"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 08:45 to 21:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "12h30", "dur_1": "12h30", "dur_3": "11h30", "dur_0": "12h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.462", "p1": "0.462", "p3": "0.500", "p0": "0.538"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 454:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "53", "age_3": "54", "age_1": "51", "age_0": "52"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = 2026-04-01; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 5 people. A shopper bought a 10 gram box and a 100 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = closed; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 14:15 to 17:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h15", "dur_3": "2h00", "dur_0": "2h45", "dur_1": "3h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p0": "0.250", "p3": "0.125", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = $635; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.286", "p3": "0.667", "p0": "0.714", "p1": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = 2026-07-05; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=934, unit_price=422, discount_pct=0, region=EU, sku_total=1356 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1356", "v3": "395082", "v0": "394148", "v2": "39414"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed October 25, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 01", "date_2": "October 28", "date_0": "October 30", "date_1": "October 30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = pending; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 550:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $818. A fee of $98 is charged, then a deposit of $47 arrives. A second fee of $49 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "715", "answer_0": "718", "answer_2": "734", "answer_1": "710"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, September 12, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, September 01, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 354:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, April 24, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = $97; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $567. A fee of $71 is charged, then a deposit of $35 arrives. A second fee of $35 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "510", "answer_2": "486", "answer_0": "496", "answer_1": "500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = $339; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 12 people. A shopper bought a 108 gram box and a 1080 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $309. A fee of $158 is charged, then a deposit of $21 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "109", "answer_0": "93", "answer_3": "116", "answer_1": "95"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p3": "0.125", "p2": "0.250", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = open; priority = high.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $805. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=374, unit_price=593, discount_pct=0, region=EU, sku_total=967 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "221782", "v1": "967", "v2": "22178", "v3": "221408"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 8 grams of salt per serving and serves 6 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1967. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "56", "age_0": "57", "age_3": "59", "age_2": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed May 15, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 30", "date_0": "May 28", "date_3": "May 26", "date_1": "May 24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p3": "0.125", "p2": "0.250", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, February 07, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p0": "0.100", "p1": "0.150", "p3": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "14", "answer_0": "24", "answer_3": "15", "answer_2": "9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 277 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "231", "answer_1": "252", "answer_0": "240", "answer_2": "229"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p3": "0.083", "p0": "0.167", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1139 per year. It increases by 15% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1275.85", "answer_2": "968.15", "answer_0": "1280.85", "answer_1": "1309.85"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=239, unit_price=402, discount_pct=0, region=EU, sku_total=641 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "641", "v2": "9607", "v0": "96078", "v3": "95839"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 220:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $809. A fee of $192 is charged, then a deposit of $77 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "583", "answer_0": "598", "answer_3": "595", "answer_1": "587"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1984. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "40", "age_0": "41", "age_3": "43", "age_2": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=679, unit_price=806, discount_pct=0, region=EU, sku_total=1485 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "54727", "v0": "547274", "v1": "1485", "v3": "546595"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 12:15 to 17:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "5h15", "dur_2": "4h45", "dur_3": "4h15", "dur_0": "4h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $433. A fee of $122 is charged, then a deposit of $21 arrives. A second fee of $61 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "271", "answer_3": "293", "answer_1": "254", "answer_2": "269"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $311. A fee of $187 is charged, then a deposit of $30 arrives. A second fee of $93 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "48", "answer_1": "91", "answer_2": "69", "answer_0": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = $605; priority = low.\n\nClaim to verify: \"the amount is $605\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $669. A fee of $171 is charged, then a deposit of $60 arrives. A second fee of $85 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "455", "answer_2": "503", "answer_0": "473", "answer_1": "482"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, January 06, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $405. A fee of $34 is charged, then a deposit of $42 arrives. A second fee of $17 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "391", "answer_2": "389", "answer_0": "396", "answer_3": "379"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 861:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1276 per year. It increases by 15% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1084.6", "answer_0": "1450.4", "answer_1": "1467.4", "answer_3": "1440.4"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed December 05, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 17", "date_0": "December 18", "date_1": "December 15", "date_2": "December 20"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 339:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.625", "p2": "0.375", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 372:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 353:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed March 25, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 07", "date_1": "April 03", "date_3": "April 05", "date_2": "April 05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.167", "p0": "0.250", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, December 12, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=784, unit_price=386, discount_pct=0, region=EU, sku_total=1170 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1170", "v0": "302624", "v3": "303408", "v2": "30262"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $987; priority = high.\n\nClaim to verify: \"the owner is $987\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=427, unit_price=262, discount_pct=0, region=EU, sku_total=689 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "112301", "v0": "111874", "v1": "689", "v2": "11187"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $265. A fee of $137 is charged, then a deposit of $64 arrives. A second fee of $68 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "146", "answer_1": "103", "answer_0": "124", "answer_2": "139"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 208:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 371:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "29", "answer_2": "25", "answer_3": "20", "answer_0": "45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p1": "0.250", "p3": "0.050", "p0": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = 2026-06-04; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p2": "0.375", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Monday, April 13, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = $845; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $471 per year. It increases by 25% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "588.75", "answer_3": "532.75", "answer_0": "561.75", "answer_2": "353.25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = 2026-09-13; priority = low.\n\nClaim to verify: \"the date is 2026-09-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Sunday, June 28, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "21", "answer_2": "12", "answer_1": "19", "answer_0": "33"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1959. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "68", "age_1": "66", "age_3": "65", "age_0": "67"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.727", "p0": "0.273", "p2": "0.727", "p3": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 560:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_2": "12", "answer_1": "19", "answer_3": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.636", "p3": "0.300", "p0": "0.364", "p2": "0.636"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $31k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "15", "answer_3": "15", "answer_1": "18", "answer_0": "30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.545", "p1": "0.455", "p3": "0.500", "p2": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 209 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "63", "answer_0": "54", "answer_2": "40", "answer_3": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 09:00 to 20:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "11h45", "dur_0": "11h15", "dur_1": "11h30", "dur_3": "10h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = 2026-02-12; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Monday, October 26, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed February 04, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 10", "date_3": "February 12", "date_1": "February 10", "date_0": "February 12"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $684. A fee of $84 is charged, then a deposit of $72 arrives. A second fee of $42 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "630", "answer_2": "603", "answer_1": "657", "answer_3": "643"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 376:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 8 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_0": "48", "answer_2": "24", "answer_1": "30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 3 grams of salt per serving and serves 9 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, July 07, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 635:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 566:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.083", "p2": "0.250", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = $18; priority = low.\n\nClaim to verify: \"the owner is $18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=768, unit_price=350, discount_pct=0, region=EU, sku_total=1118 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "26880", "v1": "1118", "v3": "269568", "v0": "268800"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p1": "0.150", "p2": "0.050", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=131, unit_price=366, discount_pct=0, region=EU, sku_total=497 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "4794", "v0": "47946", "v3": "48077", "v1": "497"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = $233; priority = low.\n\nClaim to verify: \"the amount is $233\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, February 07, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.222", "p3": "0.750", "p0": "0.778", "p2": "0.222"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $787 per year. It increases by 15% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "905.05", "answer_3": "863.05", "answer_0": "884.05", "answer_2": "668.95"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1970. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "54", "age_2": "55", "age_3": "56", "age_1": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1744 per year. It increases by 25% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1308", "answer_3": "2120", "answer_1": "2180", "answer_0": "2157"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 579:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = 2026-11-17; priority = high.\n\nClaim to verify: \"the status is 2026-11-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 383 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_3": "75", "answer_1": "90", "answer_2": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = $470; priority = high.\n\nClaim to verify: \"the status is $470\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.625", "p2": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=177, unit_price=943, discount_pct=0, region=EU, sku_total=1120 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1120", "v3": "167088", "v0": "166911", "v2": "16691"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Record fields: date = 2026-11-06; priority = high.\n\nClaim to verify: \"the date is 2026-11-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = 2026-11-22; priority = high.\n\nClaim to verify: \"the owner is 2026-11-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.286", "p3": "0.667", "p0": "0.714", "p2": "0.286"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed October 02, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "October 16", "date_1": "October 12", "date_3": "October 14", "date_2": "October 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 5 people. A shopper bought a 10 gram box and a 100 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.364", "p0": "0.636", "p3": "0.600", "p1": "0.364"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.615", "p3": "0.583", "p2": "0.385", "p1": "0.385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Wednesday, June 24, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p3": "0.083", "p0": "0.167", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 972:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = 2026-07-12; priority = low.\n\nClaim to verify: \"the amount is 2026-07-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $630. A fee of $188 is charged, then a deposit of $66 arrives. A second fee of $94 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "414", "answer_2": "408", "answer_3": "435", "answer_1": "438"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 301:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = 2026-12-14; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 5 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.545", "p1": "0.455", "p2": "0.455", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $35k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 303 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "85", "answer_2": "85", "answer_0": "100", "answer_1": "96"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = 2026-01-07; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1990. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "34", "age_2": "37", "age_0": "36", "age_1": "35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1051 per year. It increases by 20% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1245.2", "answer_3": "1276.2", "answer_1": "1261.2", "answer_2": "840.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.167", "p1": "0.500", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=391, unit_price=178, discount_pct=0, region=EU, sku_total=569 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "69989", "v2": "6959", "v1": "569", "v0": "69598"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 7 people. A shopper bought a 49 gram box and a 490 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=327, unit_price=549, discount_pct=0, region=EU, sku_total=876 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "179196", "v2": "17952", "v0": "179523", "v1": "876"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.385", "p2": "0.615", "p1": "0.615", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed October 15, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 24", "date_3": "October 24", "date_1": "October 22", "date_0": "October 26"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1966. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "60", "age_1": "58", "age_0": "59", "age_3": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, August 10, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=623, unit_price=996, discount_pct=0, region=EU, sku_total=1619 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1619", "v0": "620508", "v3": "619885", "v2": "62050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 427:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p3": "0.125", "p1": "0.500", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = 2026-12-21; priority = high.\n\nClaim to verify: \"the status is 2026-12-21\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1960. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "65", "age_0": "66", "age_2": "67", "age_3": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 328:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = $624; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 210 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "260", "answer_3": "225", "answer_0": "240", "answer_2": "251"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1963. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "62", "age_3": "63", "age_0": "61", "age_1": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p2": "0.600", "p3": "0.250", "p0": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, September 25, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 10:30 to 14:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "3h15", "dur_1": "4h15", "dur_0": "3h45", "dur_2": "4h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $680. A fee of $158 is charged, then a deposit of $32 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "492", "answer_3": "499", "answer_0": "475", "answer_2": "462"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 06:45 to 08:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "1h30", "dur_3": "1h00", "dur_1": "2h00", "dur_2": "2h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=890, unit_price=672, discount_pct=0, region=EU, sku_total=1562 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "598080", "v1": "1562", "v3": "597190", "v2": "59808"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 269:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $547. A fee of $122 is charged, then a deposit of $27 arrives. A second fee of $61 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "403", "answer_1": "420", "answer_0": "391", "answer_3": "400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1966. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "61", "age_1": "59", "age_0": "60", "age_3": "62"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.167", "p2": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed December 24, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 30", "date_0": "January 01", "date_2": "January 03", "date_3": "January 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 07:00 to 22:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "14h45", "dur_1": "15h30", "dur_2": "15h00", "dur_3": "14h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "21", "answer_3": "12", "answer_2": "9", "answer_1": "13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Thursday, October 22, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $212 per year. It increases by 20% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "207.4", "answer_0": "231.4", "answer_2": "169.6", "answer_1": "254.4"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $449. A fee of $107 is charged, then a deposit of $72 arrives. A second fee of $53 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "387", "answer_1": "384", "answer_0": "361", "answer_2": "378"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.167", "p1": "0.833", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 5 grams of salt per serving and serves 10 people. A shopper bought a 50 gram box and a 500 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = 2026-12-10; priority = high.\n\nClaim to verify: \"the date is 2026-12-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=550, unit_price=230, discount_pct=0, region=EU, sku_total=780 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "126500", "v3": "127050", "v2": "12650", "v1": "780"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1972. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "55", "age_1": "53", "age_3": "56", "age_0": "54"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 11:30 to 22:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "9h45", "dur_2": "11h00", "dur_1": "10h45", "dur_0": "10h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_0": "52", "answer_3": "28", "answer_1": "31"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, October 17, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1841 per year. It increases by 5% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1933.05", "answer_3": "1937.05", "answer_2": "1748.95", "answer_0": "1921.05"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 545:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.125", "p1": "0.375", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = 2026-09-09; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Monday, May 04, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Sunday, September 27, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=349, unit_price=535, discount_pct=0, region=EU, sku_total=884 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "884", "v2": "18671", "v0": "186715", "v3": "187064"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=433, unit_price=786, discount_pct=0, region=EU, sku_total=1219 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "340338", "v2": "34033", "v3": "340771", "v1": "1219"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1958. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "67", "age_3": "66", "age_0": "68", "age_2": "69"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.714", "p0": "0.750", "p1": "0.250", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1213 per year. It increases by 20% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1442.6", "answer_2": "970.4", "answer_1": "1455.6", "answer_3": "1450.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.462", "p1": "0.538", "p2": "0.538", "p3": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = 2026-02-17; priority = low.\n\nClaim to verify: \"the date is 2026-02-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = closed; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1932 per year. It increases by 20% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2245.4", "answer_0": "2279.4", "answer_1": "2318.4", "answer_2": "1545.6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=597, unit_price=559, discount_pct=0, region=EU, sku_total=1156 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "333126", "v1": "1156", "v0": "333723", "v2": "33372"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.800", "p3": "0.778", "p2": "0.200", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed December 02, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 09", "date_1": "December 07", "date_2": "December 07", "date_3": "December 09"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "30", "answer_2": "10", "answer_3": "20", "answer_1": "14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $648. A fee of $113 is charged, then a deposit of $73 arrives. A second fee of $56 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "571", "answer_2": "533", "answer_0": "552", "answer_3": "541"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.250", "p0": "0.375", "p1": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed October 18, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 31", "date_0": "October 29", "date_3": "October 29", "date_1": "October 27"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.222", "p1": "0.778", "p2": "0.778"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 338 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "168", "answer_3": "161", "answer_2": "154", "answer_1": "151"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $624. A fee of $105 is charged, then a deposit of $28 arrives. A second fee of $52 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "516", "answer_2": "515", "answer_0": "495", "answer_1": "497"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = 2026-01-07; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "16", "answer_0": "24", "answer_2": "8", "answer_1": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=200, unit_price=623, discount_pct=0, region=EU, sku_total=823 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "124800", "v2": "12460", "v1": "823", "v0": "124600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, November 05, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "48", "answer_1": "54", "answer_3": "48", "answer_0": "96"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_0": "35", "answer_1": "23", "answer_2": "21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 6 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Sunday, November 15, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $40k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $20k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed June 24, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 02", "date_3": "July 04", "date_0": "July 06", "date_2": "July 04"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "49", "answer_2": "14", "answer_3": "35", "answer_1": "19"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.571", "p1": "0.375", "p2": "0.375", "p0": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = 2026-11-20; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p1": "0.200", "p2": "0.100", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, January 03, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1971. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "55", "age_3": "52", "age_0": "54", "age_1": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1958. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "70", "age_1": "67", "age_0": "68", "age_2": "69"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p2": "0.375", "p3": "0.125", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=401, unit_price=770, discount_pct=0, region=EU, sku_total=1171 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1171", "v2": "30877", "v3": "309171", "v0": "308770"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed August 17, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "August 31", "date_2": "August 29", "date_1": "August 27", "date_3": "August 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.667", "p2": "0.500", "p1": "0.833"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.600", "p1": "0.400", "p2": "0.400", "p3": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed June 25, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 03", "date_1": "July 01", "date_2": "July 01", "date_3": "July 03"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = $95; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $750. A fee of $103 is charged, then a deposit of $58 arrives. A second fee of $51 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "684", "answer_2": "637", "answer_0": "654", "answer_1": "670"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, July 05, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $860 per year. It increases by 5% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "918", "answer_2": "817", "answer_0": "891", "answer_1": "903"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, February 03, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = 2026-08-14; priority = low.\n\nClaim to verify: \"the amount is 2026-08-14\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = $403; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 3 grams of salt per serving and serves 8 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = open; priority = high.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = $725; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, June 24, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, May 19, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 11:45 to 20:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "8h30", "dur_1": "9h00", "dur_2": "9h00", "dur_3": "8h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Saturday, April 18, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1770 per year. It increases by 25% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2212.5", "answer_0": "2184.5", "answer_3": "2189.5", "answer_2": "1327.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.250", "p1": "0.417", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1965. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "59", "age_2": "61", "age_3": "62", "age_0": "60"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 320:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 08:45 to 21:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "12h45", "dur_1": "12h45", "dur_3": "11h45", "dur_0": "12h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.625", "p3": "0.286", "p0": "0.375", "p1": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $880. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 191 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "171", "answer_3": "154", "answer_0": "160", "answer_2": "140"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.333", "p3": "0.167", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 218 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "107", "answer_1": "108", "answer_3": "114", "answer_0": "96"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $449. A fee of $83 is charged, then a deposit of $48 arrives. A second fee of $41 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "373", "answer_3": "363", "answer_2": "368", "answer_1": "368"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 7 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Saturday, February 21, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 9 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 6 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=579, unit_price=357, discount_pct=0, region=EU, sku_total=936 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "936", "v2": "20670", "v3": "207282", "v0": "206703"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "70", "answer_1": "46", "answer_2": "42", "answer_3": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 288:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A shift runs from 06:15 to 10:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "3h15", "dur_2": "3h30", "dur_3": "3h00", "dur_1": "4h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "48", "answer_0": "64", "answer_1": "50", "answer_3": "16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.167", "p0": "0.333", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 78 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "43", "answer_0": "48", "answer_1": "59", "answer_2": "52"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.600", "p1": "0.600", "p3": "0.250", "p0": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p2": "0.150", "p3": "0.050", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 114 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "210", "answer_3": "226", "answer_1": "229", "answer_2": "193"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 770:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1966. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "59", "age_3": "57", "age_2": "60", "age_1": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = 2026-08-16; priority = low.\n\nClaim to verify: \"the amount is 2026-08-16\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=239, unit_price=133, discount_pct=0, region=EU, sku_total=372 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "32026", "v0": "31787", "v1": "372", "v2": "3178"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = $287; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 342:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = $420; priority = high.\n\nClaim to verify: \"the owner is $420\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 894:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.167", "p1": "0.833", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 439:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 481:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 216:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.714", "p0": "0.750", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.500", "p3": "0.167", "p1": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.750", "p2": "0.222", "p0": "0.778", "p1": "0.222"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "65", "answer_3": "35", "answer_2": "30", "answer_1": "37"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $895. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $39k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $666 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "632.7", "answer_1": "699.3", "answer_3": "674.3", "answer_0": "667.3"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $680. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.375", "p3": "0.125", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Wednesday, February 18, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.583", "p0": "0.615", "p1": "0.385", "p2": "0.385"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=365, unit_price=417, discount_pct=0, region=EU, sku_total=782 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "15220", "v1": "782", "v3": "151840", "v0": "152205"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed November 15, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 22", "date_1": "November 22", "date_3": "November 24", "date_0": "November 24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $761 per year. It increases by 20% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "608.8", "answer_1": "913.2", "answer_3": "854.2", "answer_0": "889.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 12 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.200", "p2": "0.667", "p0": "0.333", "p1": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 377 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "137", "answer_0": "144", "answer_2": "135", "answer_3": "147"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 965:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, June 05, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=327, unit_price=209, discount_pct=0, region=EU, sku_total=536 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "536", "v3": "68670", "v2": "6834", "v0": "68343"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 257 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_2": "39", "answer_3": "39", "answer_1": "62"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1961. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "65", "age_2": "66", "age_3": "63", "age_1": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 18:30 to 19:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h30", "dur_2": "0h30", "dur_1": "0h30", "dur_0": "0h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.250", "p1": "0.500", "p0": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.250", "p2": "0.125", "p1": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.500", "p2": "0.167", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 6 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 142 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "114", "answer_3": "102", "answer_0": "96", "answer_1": "109"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 307 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "216", "answer_1": "235", "answer_2": "211", "answer_3": "235"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=800, unit_price=579, discount_pct=0, region=EU, sku_total=1379 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "46320", "v0": "463200", "v3": "464000", "v1": "1379"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.222", "p3": "0.750", "p1": "0.222", "p0": "0.778"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = 2026-09-24; priority = low.\n\nClaim to verify: \"the date is 2026-09-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed June 20, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 27", "date_1": "June 25", "date_0": "June 26", "date_2": "June 24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1978. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "48", "age_1": "45", "age_2": "47", "age_0": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.444", "p2": "0.444", "p0": "0.556", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed August 27, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 02", "date_3": "September 02", "date_2": "August 31", "date_1": "August 31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 4 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = open; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, December 27, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 5 people. A shopper bought a 25 gram box and a 250 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $846. A fee of $153 is charged, then a deposit of $75 arrives. A second fee of $76 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "711", "answer_0": "692", "answer_1": "677", "answer_3": "694"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = 2026-09-17; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed January 07, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 16", "date_0": "January 16", "date_1": "January 14", "date_2": "January 18"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = ops; priority = low.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 3 grams of salt per serving and serves 12 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 9 grams of salt per serving and serves 8 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $772. A fee of $70 is charged, then a deposit of $20 arrives. A second fee of $35 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "666", "answer_3": "664", "answer_2": "700", "answer_0": "687"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 429:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.400", "p0": "0.600", "p1": "0.400", "p3": "0.556"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, December 10, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed October 10, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 13", "date_3": "October 15", "date_0": "October 14", "date_2": "October 12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed May 04, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 15", "date_1": "May 13", "date_2": "May 13", "date_3": "May 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 508:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 239 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "90", "answer_2": "86", "answer_0": "80", "answer_1": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p0": "0.333", "p2": "0.667", "p1": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1960. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "66", "age_1": "65", "age_3": "64", "age_2": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.167", "p1": "0.333", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p2": "0.375", "p0": "0.500", "p3": "0.125"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 490:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=246, unit_price=492, discount_pct=0, region=EU, sku_total=738 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "121278", "v0": "121032", "v1": "738", "v2": "12103"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $684. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=517, unit_price=794, discount_pct=0, region=EU, sku_total=1311 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "410498", "v1": "1311", "v3": "411015", "v2": "41049"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.600", "p3": "0.500", "p2": "0.400", "p1": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "36", "answer_1": "39", "answer_3": "18", "answer_0": "54"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=355, unit_price=462, discount_pct=0, region=EU, sku_total=817 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "163655", "v0": "164010", "v1": "817", "v2": "16401"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $205. A fee of $185 is charged, then a deposit of $46 arrives. A second fee of $92 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "-24", "answer_2": "-42", "answer_0": "-26", "answer_3": "4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "30", "answer_0": "50", "answer_2": "25", "answer_3": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, August 24, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 305:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_3": "30", "answer_0": "60", "answer_1": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Saturday, November 21, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=407, unit_price=807, discount_pct=0, region=EU, sku_total=1214 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1214", "v3": "328042", "v2": "32844", "v0": "328449"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $842. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = $259; priority = high.\n\nClaim to verify: \"the status is $259\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1971. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "54", "age_1": "52", "age_0": "53", "age_3": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 17:30 to 17:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "0h00", "dur_3": "-1h15", "dur_1": "0h15", "dur_0": "-1h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = 2026-01-01; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $599. A fee of $40 is charged, then a deposit of $45 arrives. A second fee of $20 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "594", "answer_0": "584", "answer_2": "594", "answer_1": "567"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=787, unit_price=216, discount_pct=0, region=EU, sku_total=1003 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "169992", "v2": "16999", "v3": "170779", "v1": "1003"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed November 23, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 02", "date_2": "November 30", "date_0": "December 02", "date_1": "November 30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 638:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.250", "p2": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = $448; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed May 24, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 01", "date_1": "May 30", "date_0": "June 01", "date_2": "June 03"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed October 26, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 03", "date_0": "November 03", "date_1": "November 01", "date_2": "November 05"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = $888; priority = high.\n\nClaim to verify: \"the owner is $888\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = $753; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.500", "p1": "0.625", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1967. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "59", "age_1": "58", "age_2": "60", "age_3": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 5 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $429. A fee of $158 is charged, then a deposit of $26 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "191", "answer_1": "197", "answer_3": "196", "answer_0": "218"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = $157; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.333", "p1": "0.417", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 4 people. A shopper bought a 8 gram box and a 80 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.250", "p0": "0.375", "p3": "0.125"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $356. A fee of $181 is charged, then a deposit of $27 arrives. A second fee of $90 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "142", "answer_2": "142", "answer_0": "112", "answer_1": "106"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $444. A fee of $167 is charged, then a deposit of $44 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "218", "answer_3": "268", "answer_2": "246", "answer_0": "238"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 160 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "79", "answer_0": "81", "answer_1": "76", "answer_3": "79"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.125", "p1": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed May 05, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 08", "date_2": "May 10", "date_3": "May 10", "date_0": "May 08"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "28", "answer_1": "25", "answer_0": "49", "answer_2": "21"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1009 per year. It increases by 10% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "908.1", "answer_0": "1067.9", "answer_1": "1109.9", "answer_3": "1055.9"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.333", "p3": "0.636", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = 2026-09-12; priority = high.\n\nClaim to verify: \"the status is 2026-09-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = $20; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $285 per year. It increases by 15% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "242.25", "answer_0": "289.75", "answer_3": "280.75", "answer_1": "327.75"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=850, unit_price=411, discount_pct=0, region=EU, sku_total=1261 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "349350", "v1": "1261", "v2": "34935", "v3": "348500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = 2026-12-02; priority = low.\n\nClaim to verify: \"the owner is 2026-12-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $554. A fee of $49 is charged, then a deposit of $83 arrives. A second fee of $24 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "564", "answer_3": "583", "answer_1": "552", "answer_2": "576"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.250", "p0": "0.333", "p1": "0.417"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=114, unit_price=186, discount_pct=0, region=EU, sku_total=300 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "2120", "v1": "300", "v0": "21204", "v3": "21090"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = 2026-10-12; priority = low.\n\nClaim to verify: \"the status is 2026-10-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=676, unit_price=140, discount_pct=0, region=EU, sku_total=816 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "93964", "v1": "816", "v2": "9464", "v0": "94640"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 904:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1967. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "55", "age_1": "56", "age_0": "57", "age_2": "58"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = $868; priority = high.\n\nClaim to verify: \"the status is $868\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1193 per year. It increases by 5% this year, then a $40 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1195.65", "answer_0": "1212.65", "answer_2": "1133.35", "answer_1": "1252.65"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.667", "p3": "0.167", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, April 16, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = $943; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p0": "0.375", "p3": "0.286", "p2": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 523:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Form fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Wednesday, May 20, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, May 26, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.500", "p3": "0.444", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed November 23, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 28", "date_3": "November 30", "date_2": "November 28", "date_0": "November 30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "35", "answer_3": "30", "answer_0": "60", "answer_2": "30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 134 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "64", "answer_0": "72", "answer_2": "84", "answer_3": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $801. A fee of $60 is charged, then a deposit of $24 arrives. A second fee of $30 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "731", "answer_1": "765", "answer_2": "723", "answer_0": "735"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed November 19, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 24", "date_3": "November 24", "date_2": "November 26", "date_1": "November 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = $204; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = 2026-12-17; priority = high.\n\nClaim to verify: \"the owner is 2026-12-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 233:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p3": "0.050", "p0": "0.150", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $209. A fee of $91 is charged, then a deposit of $75 arrives. A second fee of $45 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "157", "answer_0": "148", "answer_1": "157", "answer_3": "133"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.083", "p0": "0.333", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $956 per year. It increases by 20% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1118.2", "answer_2": "764.8", "answer_1": "1147.2", "answer_3": "1134.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 17:45 to 21:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "3h45", "dur_1": "4h00", "dur_3": "3h00", "dur_2": "4h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 115 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "67", "answer_2": "57", "answer_3": "73", "answer_0": "64"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1765 per year. It increases by 25% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1323.75", "answer_1": "2206.25", "answer_0": "2183.25", "answer_3": "2188.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1966. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_3": "60", "age_2": "59", "age_1": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $353. A fee of $170 is charged, then a deposit of $21 arrives. A second fee of $85 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "119", "answer_3": "111", "answer_2": "122", "answer_1": "96"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, March 15, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $343 per year. It increases by 5% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "325.85", "answer_0": "349.15", "answer_3": "310.15", "answer_1": "360.15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.417", "p1": "0.417", "p0": "0.583", "p3": "0.545"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1258 per year. It increases by 15% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1447.7", "answer_2": "1069.3", "answer_0": "1410.7", "answer_1": "1446.7"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $585. A fee of $125 is charged, then a deposit of $52 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "434", "answer_1": "427", "answer_2": "452", "answer_0": "450"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 715:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "22", "answer_2": "15", "answer_0": "50", "answer_3": "35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.150", "p2": "0.050", "p0": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = 2026-03-05; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 307:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $585. A fee of $33 is charged, then a deposit of $42 arrives. A second fee of $16 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "556", "answer_2": "597", "answer_0": "578", "answer_1": "596"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=896, unit_price=862, discount_pct=0, region=EU, sku_total=1758 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "773248", "v1": "1758", "v2": "77235", "v0": "772352"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = 2026-08-06; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $546. A fee of $167 is charged, then a deposit of $44 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "332", "answer_1": "315", "answer_0": "340", "answer_3": "361"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p1": "0.667", "p0": "0.333", "p2": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An account starts at $524. A fee of $30 is charged, then a deposit of $72 arrives. A second fee of $15 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "551", "answer_2": "521", "answer_1": "578", "answer_3": "549"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Saturday, May 16, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=550, unit_price=417, discount_pct=0, region=EU, sku_total=967 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "22935", "v1": "967", "v0": "229350", "v3": "229900"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 12:15 to 14:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "2h00", "dur_0": "1h30", "dur_1": "2h15", "dur_3": "1h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 346:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 8 grams of salt per serving and serves 10 people. A shopper bought a 80 gram box and a 800 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = 2026-06-15; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $281. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p3": "0.778", "p0": "0.800", "p2": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $960 per year. It increases by 20% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1152", "answer_2": "768", "answer_0": "1108", "answer_3": "1132"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $335 per year. It increases by 10% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "351.5", "answer_0": "333.5", "answer_1": "368.5", "answer_2": "301.5"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 111 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "100", "answer_0": "108", "answer_2": "96", "answer_3": "94"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, December 23, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed December 11, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 22", "date_1": "December 18", "date_2": "December 20", "date_3": "December 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "25", "answer_3": "10", "answer_0": "35", "answer_1": "27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1966. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "58", "age_2": "61", "age_1": "59", "age_0": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $9k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = closed; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 318 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "77", "answer_1": "95", "answer_0": "75", "answer_2": "91"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $25k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1699 per year. It increases by 10% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1827.9", "answer_1": "1868.9", "answer_2": "1529.1", "answer_0": "1833.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $926 per year. It increases by 10% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1018.6", "answer_3": "985.6", "answer_0": "1006.6", "answer_2": "833.4"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $357. A fee of $120 is charged, then a deposit of $56 arrives. A second fee of $60 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "252", "answer_0": "233", "answer_1": "203", "answer_2": "242"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Friday, April 17, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p3": "0.083", "p0": "0.167", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $184. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = 2026-05-23; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1972. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "55", "age_0": "53", "age_1": "52", "age_2": "54"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p3": "0.167", "p0": "0.667", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 259:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_1": "24", "answer_2": "20", "answer_3": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1979. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "44", "age_1": "45", "age_0": "46", "age_2": "47"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 28, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 03", "date_0": "May 01", "date_2": "April 29", "date_1": "May 01"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1532 per year. It increases by 10% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1668.2", "answer_3": "1640.2", "answer_2": "1378.8", "answer_1": "1685.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $730 per year. It increases by 5% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "764.5", "answer_1": "766.5", "answer_0": "743.5", "answer_2": "693.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_1": "23", "answer_3": "30", "answer_2": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed September 27, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 30", "date_1": "September 30", "date_2": "September 28", "date_3": "October 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1121 per year. It increases by 5% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1177.05", "answer_2": "1064.95", "answer_0": "1161.05", "answer_3": "1156.05"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1972. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "51", "age_3": "54", "age_2": "53", "age_0": "52"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = 2026-11-11; priority = low.\n\nClaim to verify: \"the status is 2026-11-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 08:30 to 19:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "11h00", "dur_1": "11h15", "dur_2": "11h30", "dur_3": "10h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=767, unit_price=347, discount_pct=0, region=EU, sku_total=1114 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "26614", "v0": "266149", "v3": "265382", "v1": "1114"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Contract signed January 25, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 04", "date_0": "February 02", "date_3": "February 02", "date_1": "January 31"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p2": "0.500", "p3": "0.167", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_2": "12", "answer_1": "14", "answer_0": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 8 grams of salt per serving and serves 11 people. A shopper bought a 88 gram box and a 880 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.167", "p0": "0.250", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = $132; priority = high.\n\nClaim to verify: \"the owner is $132\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 266 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "67", "answer_3": "56", "answer_1": "85", "answer_0": "72"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 229 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "196", "answer_1": "182", "answer_2": "179", "answer_3": "212"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed November 12, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 21", "date_0": "November 23", "date_1": "November 19", "date_2": "November 25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = 2026-05-28; priority = high.\n\nClaim to verify: \"the status is 2026-05-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = $163; priority = low.\n\nClaim to verify: \"the date is $163\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $286 per year. It increases by 5% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "217.3", "answer_1": "300.3", "answer_0": "252.3", "answer_2": "271.7"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.150", "p2": "0.050", "p0": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $21k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=247, unit_price=415, discount_pct=0, region=EU, sku_total=662 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "10250", "v0": "102505", "v1": "662", "v3": "102752"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.444", "p2": "0.556", "p1": "0.556", "p3": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 864:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, September 15, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=518, unit_price=383, discount_pct=0, region=EU, sku_total=901 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "197876", "v1": "901", "v0": "198394", "v2": "19839"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = open; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "40", "answer_0": "88", "answer_1": "46", "answer_3": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.462", "p1": "0.538", "p3": "0.417", "p2": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Tuesday, July 28, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_1": "26", "answer_0": "56", "answer_2": "21"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $512 per year. It increases by 25% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "599", "answer_3": "610", "answer_2": "384", "answer_1": "640"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 321 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_1": "74", "answer_3": "67", "answer_2": "69"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.556", "p1": "0.556", "p3": "0.375", "p0": "0.444"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = $71; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 672:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.667", "p0": "0.333", "p3": "0.273"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, August 10, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = 2026-07-09; priority = low.\n\nClaim to verify: \"the date is 2026-07-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $761. A fee of $47 is charged, then a deposit of $46 arrives. A second fee of $23 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "737", "answer_2": "730", "answer_3": "714", "answer_1": "741"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, November 15, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $672. A fee of $56 is charged, then a deposit of $55 arrives. A second fee of $28 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "660", "answer_3": "657", "answer_1": "663", "answer_0": "643"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=165, unit_price=665, discount_pct=0, region=EU, sku_total=830 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "109725", "v2": "10972", "v3": "109560", "v1": "830"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 18:00 to 19:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "1h00", "dur_3": "0h00", "dur_0": "0h15", "dur_2": "0h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1990. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "35", "age_0": "34", "age_3": "32", "age_1": "33"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 12:45 to 15:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h15", "dur_0": "1h45", "dur_3": "1h15", "dur_2": "2h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.333", "p3": "0.167", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.545", "p1": "0.455", "p2": "0.455"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = 2026-08-14; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = 2026-06-18; priority = low.\n\nClaim to verify: \"the owner is 2026-06-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = $30; priority = high.\n\nClaim to verify: \"the date is $30\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1021 per year. It increases by 20% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1195.2", "answer_1": "1225.2", "answer_3": "1235.2", "answer_2": "816.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1969. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "55", "age_0": "56", "age_2": "57", "age_3": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, December 09, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed December 12, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 24", "date_2": "December 27", "date_1": "December 22", "date_0": "December 25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $735. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.250", "p1": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = pending; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = pending; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 896:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.750", "p2": "0.222", "p1": "0.222", "p0": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.200", "p1": "0.667", "p2": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $347. A fee of $130 is charged, then a deposit of $74 arrives. A second fee of $65 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "246", "answer_0": "226", "answer_2": "232", "answer_3": "252"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.750", "p2": "0.750", "p0": "0.250", "p3": "0.143"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = $203; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p3": "0.167", "p0": "0.667", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1972. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "54", "age_2": "53", "age_1": "51", "age_0": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = 2026-04-20; priority = low.\n\nClaim to verify: \"the status is 2026-04-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.150", "p1": "0.250", "p0": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 327 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "50", "answer_1": "34", "answer_2": "35", "answer_3": "33"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "14", "answer_3": "42", "answer_0": "56", "answer_1": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p2": "0.667", "p1": "0.667", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=433, unit_price=613, discount_pct=0, region=EU, sku_total=1046 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "265862", "v2": "26542", "v0": "265429", "v1": "1046"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 241 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "164", "answer_2": "171", "answer_0": "168", "answer_1": "174"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 16:30 to 17:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "0h00", "dur_3": "-1h30", "dur_0": "-1h45", "dur_1": "0h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = ops; priority = low.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=696, unit_price=628, discount_pct=0, region=EU, sku_total=1324 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1324", "v0": "437088", "v3": "436392", "v2": "43708"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $873. A fee of $180 is charged, then a deposit of $48 arrives. A second fee of $90 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "679", "answer_0": "651", "answer_1": "657", "answer_3": "634"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "27", "answer_2": "24", "answer_3": "12", "answer_0": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 73 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "364", "answer_1": "364", "answer_2": "370", "answer_0": "378"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, August 25, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.444", "p2": "0.444", "p0": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed November 14, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 19", "date_2": "November 17", "date_3": "November 20", "date_1": "November 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = $47; priority = high.\n\nClaim to verify: \"the status is $47\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=964, unit_price=880, discount_pct=0, region=EU, sku_total=1844 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "847356", "v0": "848320", "v2": "84832", "v1": "1844"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p3": "0.167", "p2": "0.500", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 723:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 8 grams of salt per serving and serves 6 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = closed; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, May 12, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $35k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=272, unit_price=732, discount_pct=0, region=EU, sku_total=1004 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1004", "v0": "199104", "v2": "19910", "v3": "199376"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, February 04, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = 2026-05-03; priority = high.\n\nClaim to verify: \"the date is 2026-05-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed June 27, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 05", "date_1": "July 02", "date_3": "July 04", "date_0": "July 03"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.083", "p2": "0.250", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: date = $824; priority = high.\n\nClaim to verify: \"the date is $824\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p2": "0.200", "p0": "0.800", "p3": "0.778"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_2": "15", "answer_1": "20", "answer_3": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.500", "p3": "0.444", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.364", "p3": "0.300", "p2": "0.636", "p1": "0.636"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Sunday, November 15, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $726 per year. It increases by 5% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "778.3", "answer_2": "689.7", "answer_1": "762.3", "answer_0": "744.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $873. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=383, unit_price=316, discount_pct=0, region=EU, sku_total=699 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "121411", "v1": "699", "v2": "12102", "v0": "121028"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed December 28, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 04", "date_0": "January 06", "date_3": "January 06", "date_2": "January 04"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 235 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "126", "answer_1": "107", "answer_3": "130", "answer_2": "116"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "32", "answer_2": "8", "answer_3": "24", "answer_1": "14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1990. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "35", "age_0": "36", "age_2": "37", "age_3": "34"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 8 grams of salt per serving and serves 11 people. A shopper bought a 88 gram box and a 880 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = $196; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "20", "answer_2": "18", "answer_0": "30", "answer_3": "12"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $543. A fee of $134 is charged, then a deposit of $75 arrives. A second fee of $67 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "399", "answer_0": "417", "answer_1": "404", "answer_2": "447"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = 2026-06-11; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $14k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "51", "answer_2": "48", "answer_3": "24", "answer_0": "72"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 962:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.538", "p3": "0.417", "p0": "0.462", "p1": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "30", "answer_3": "21", "answer_1": "16", "answer_2": "9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = $600; priority = high.\n\nClaim to verify: \"the amount is $600\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p1": "0.250", "p3": "0.050", "p0": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.375", "p3": "0.125", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 249 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "19", "answer_3": "47", "answer_2": "42", "answer_0": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = $271; priority = high.\n\nClaim to verify: \"the owner is $271\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = pending; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_1": "30", "answer_3": "48", "answer_0": "72"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 242 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_2": "50", "answer_1": "49", "answer_0": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $726. A fee of $75 is charged, then a deposit of $45 arrives. A second fee of $37 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "654", "answer_2": "672", "answer_0": "659", "answer_3": "656"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed February 08, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 20", "date_1": "February 18", "date_0": "February 20", "date_2": "February 18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $472. A fee of $87 is charged, then a deposit of $22 arrives. A second fee of $43 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "339", "answer_1": "366", "answer_3": "374", "answer_0": "364"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.538", "p3": "0.417", "p2": "0.538", "p0": "0.462"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.533", "p0": "0.467", "p3": "0.429", "p2": "0.533"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1467 per year. It increases by 10% this year, then a $49 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1613.7", "answer_0": "1564.7", "answer_3": "1543.7", "answer_2": "1320.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 645:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, January 17, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.545", "p1": "0.455", "p3": "0.500", "p2": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 3 grams of salt per serving and serves 6 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1986 per year. It increases by 5% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2106.3", "answer_0": "2066.3", "answer_1": "2085.3", "answer_2": "1886.7"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.625", "p3": "0.125", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "19", "answer_3": "20", "answer_2": "15", "answer_0": "35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $377. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed February 20, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 27", "date_3": "February 25", "date_1": "February 23", "date_0": "February 25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 17:00 to 18:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "0h30", "dur_2": "1h15", "dur_0": "0h45", "dur_1": "1h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = $126; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_2": "42", "answer_1": "44", "answer_0": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p1": "0.250", "p0": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = 2026-05-15; priority = low.\n\nClaim to verify: \"the date is 2026-05-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.417", "p1": "0.417", "p3": "0.545", "p0": "0.583"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1826 per year. It increases by 20% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2165.2", "answer_2": "1460.8", "answer_3": "2146.2", "answer_1": "2191.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 382 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "36", "answer_2": "61", "answer_3": "57", "answer_0": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1657 per year. It increases by 10% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1794.7", "answer_3": "1828.7", "answer_2": "1491.3", "answer_1": "1822.7"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $842. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=190, unit_price=737, discount_pct=0, region=EU, sku_total=927 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "140220", "v2": "14003", "v1": "927", "v0": "140030"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.615", "p1": "0.615", "p0": "0.385", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "18", "answer_0": "48", "answer_3": "36", "answer_2": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $569. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $863. A fee of $181 is charged, then a deposit of $64 arrives. A second fee of $90 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "635", "answer_0": "656", "answer_1": "652", "answer_3": "645"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = 2026-10-20; priority = low.\n\nClaim to verify: \"the owner is 2026-10-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = $494; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.625", "p3": "0.125", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_2": "32", "answer_1": "35", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1969. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "54", "age_3": "57", "age_0": "55", "age_2": "56"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, July 20, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=157, unit_price=795, discount_pct=0, region=EU, sku_total=952 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "124658", "v2": "12481", "v1": "952", "v0": "124815"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $98. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 111 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "46", "answer_0": "48", "answer_2": "50", "answer_3": "31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $899. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 336 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "46", "answer_1": "74", "answer_2": "50", "answer_0": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 4 people. A shopper bought a 8 gram box and a 80 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1505 per year. It increases by 10% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1620.5", "answer_1": "1655.5", "answer_3": "1610.5", "answer_2": "1354.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_1": "22", "answer_2": "16", "answer_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $239 per year. It increases by 10% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "226.9", "answer_3": "207.9", "answer_1": "262.9", "answer_2": "215.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = ops; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 178 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "26", "answer_1": "15", "answer_3": "32", "answer_0": "18"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $427. A fee of $133 is charged, then a deposit of $47 arrives. A second fee of $66 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "275", "answer_2": "263", "answer_3": "249", "answer_1": "247"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 400 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "119", "answer_1": "116", "answer_2": "131", "answer_0": "126"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = $105; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed June 27, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 07", "date_3": "July 09", "date_0": "July 10", "date_2": "July 12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 836:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1975. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "52", "age_0": "51", "age_1": "50", "age_3": "53"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 606:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 119 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "228", "answer_3": "231", "answer_0": "240", "answer_2": "228"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.500", "p1": "0.833", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=721, unit_price=241, discount_pct=0, region=EU, sku_total=962 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "962", "v3": "174482", "v2": "17376", "v0": "173761"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1983. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "43", "age_3": "41", "age_2": "44", "age_1": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = 2026-11-07; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=181, unit_price=903, discount_pct=0, region=EU, sku_total=1084 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "163443", "v2": "16344", "v1": "1084", "v3": "163624"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $272. A fee of $41 is charged, then a deposit of $33 arrives. A second fee of $20 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "244", "answer_2": "260", "answer_3": "258", "answer_1": "225"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed August 25, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 04", "date_0": "September 08", "date_3": "September 06", "date_2": "September 06"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $693. A fee of $89 is charged, then a deposit of $61 arrives. A second fee of $44 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "591", "answer_1": "643", "answer_0": "621", "answer_3": "592"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "22", "answer_2": "18", "answer_0": "30", "answer_3": "12"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.727", "p3": "0.200", "p0": "0.273", "p2": "0.727"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p3": "0.467", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1963. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "63", "age_0": "62", "age_3": "64", "age_1": "61"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p1": "0.833", "p2": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = $966; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed June 12, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 23", "date_0": "June 25", "date_3": "June 23", "date_1": "June 21"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 393:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "30", "answer_1": "15", "answer_3": "18", "answer_2": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed February 15, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 24", "date_1": "February 22", "date_3": "February 24", "date_2": "February 26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $4k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = 2026-10-13; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $511. A fee of $142 is charged, then a deposit of $86 arrives. A second fee of $71 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "384", "answer_3": "376", "answer_1": "373", "answer_2": "358"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $889. A fee of $171 is charged, then a deposit of $89 arrives. A second fee of $85 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "749", "answer_1": "712", "answer_2": "693", "answer_0": "722"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed March 21, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 31", "date_0": "April 03", "date_2": "April 01", "date_3": "April 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "55", "answer_2": "48", "answer_0": "104", "answer_3": "56"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p3": "0.273", "p2": "0.667", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1974. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "50", "age_0": "52", "age_2": "53", "age_1": "51"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed February 17, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 28", "date_0": "March 02", "date_3": "February 28", "date_1": "February 26"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.533", "p1": "0.533", "p3": "0.429", "p0": "0.467"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $63. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 283 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "140", "answer_1": "131", "answer_3": "128", "answer_2": "134"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 893:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.385", "p0": "0.615", "p2": "0.385", "p3": "0.583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 328 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "77", "answer_0": "81", "answer_3": "101", "answer_1": "99"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1460 per year. It increases by 5% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1491", "answer_2": "1387", "answer_3": "1457", "answer_1": "1533"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $876. A fee of $149 is charged, then a deposit of $85 arrives. A second fee of $74 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "738", "answer_1": "712", "answer_2": "725", "answer_3": "765"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, September 06, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1973. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "51", "age_1": "50", "age_2": "52", "age_3": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 378 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "93", "answer_3": "97", "answer_1": "90", "answer_0": "100"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.333", "p3": "0.273", "p2": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = open; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1967 per year. It increases by 5% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2065.35", "answer_2": "1868.65", "answer_0": "2049.35", "answer_3": "2077.35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = 2026-01-17; priority = low.\n\nClaim to verify: \"the status is 2026-01-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p1": "0.250", "p3": "0.083", "p2": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $645. A fee of $142 is charged, then a deposit of $44 arrives. A second fee of $71 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "454", "answer_1": "503", "answer_2": "446", "answer_0": "476"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p0": "0.333", "p3": "0.083", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1969. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "58", "age_2": "57", "age_1": "55", "age_0": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = 2026-05-16; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1970. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "53", "age_0": "54", "age_2": "55", "age_3": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_0": "50", "answer_2": "15", "answer_1": "22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1591 per year. It increases by 25% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1923.75", "answer_2": "1193.25", "answer_0": "1961.75", "answer_1": "1988.75"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.727", "p1": "0.727", "p0": "0.273", "p3": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 10 people. A shopper bought a 70 gram box and a 700 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $26k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 293:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 172 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "91", "answer_1": "92", "answer_0": "96", "answer_2": "92"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = 2026-09-15; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed October 11, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 17", "date_1": "October 15", "date_2": "October 17", "date_0": "October 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.125", "p1": "0.375", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.538", "p2": "0.429", "p0": "0.571", "p1": "0.429"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 331 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "39", "answer_0": "24", "answer_3": "18", "answer_1": "5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1987. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "37", "age_2": "39", "age_0": "38", "age_3": "40"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $735 per year. It increases by 5% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "771.75", "answer_0": "743.75", "answer_3": "708.75", "answer_2": "698.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $356 per year. It increases by 10% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "357.6", "answer_1": "391.6", "answer_3": "340.6", "answer_2": "320.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "56", "answer_2": "24", "answer_0": "80", "answer_1": "31"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $13k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p2": "0.050", "p3": "0.050", "p1": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 17:15 to 22:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "5h00", "dur_3": "4h15", "dur_2": "5h30", "dur_1": "5h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $844. A fee of $41 is charged, then a deposit of $65 arrives. A second fee of $20 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "825", "answer_3": "839", "answer_1": "855", "answer_0": "848"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.625", "p3": "0.286", "p1": "0.625", "p0": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.700", "p3": "0.222", "p0": "0.300", "p1": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 359 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "23", "answer_1": "17", "answer_0": "36", "answer_3": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 05, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 13", "date_0": "April 15", "date_3": "April 15", "date_1": "April 13"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p1": "0.250", "p3": "0.050", "p2": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = $292; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $216. A fee of $82 is charged, then a deposit of $35 arrives. A second fee of $41 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "125", "answer_0": "128", "answer_2": "113", "answer_3": "156"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, February 16, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = ops; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 370:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 137 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "60", "answer_0": "64", "answer_2": "71", "answer_3": "69"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Sunday, November 01, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $40k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.300", "p0": "0.700", "p3": "0.667", "p2": "0.300"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = closed; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=533, unit_price=670, discount_pct=0, region=EU, sku_total=1203 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "357643", "v1": "1203", "v0": "357110", "v2": "35711"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 356 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "99", "answer_0": "96", "answer_2": "85", "answer_1": "80"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = 2026-03-25; priority = high.\n\nClaim to verify: \"the date is 2026-03-25\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "35", "answer_2": "30", "answer_3": "25", "answer_0": "55"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed January 25, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 06", "date_1": "February 02", "date_3": "February 04", "date_0": "February 04"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 6 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 5 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Wednesday, August 05, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = $885; priority = high.\n\nClaim to verify: \"the owner is $885\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = $523; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "15", "answer_1": "22", "answer_3": "21", "answer_0": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = 2026-12-05; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $21k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "30", "answer_1": "18", "answer_2": "12", "answer_3": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 12:15 to 13:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h15", "dur_3": "0h15", "dur_0": "0h45", "dur_1": "1h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1444 per year. It increases by 5% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1486.2", "answer_0": "1493.2", "answer_1": "1516.2", "answer_2": "1371.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = $376; priority = high.\n\nClaim to verify: \"the date is $376\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 224 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "30", "answer_2": "17", "answer_0": "24", "answer_3": "7"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.583", "p2": "0.385", "p0": "0.615", "p1": "0.385"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = $908; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 352:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "24", "answer_3": "20", "answer_2": "20", "answer_0": "40"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1218 per year. It increases by 20% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "974.4", "answer_1": "1461.6", "answer_3": "1433.6", "answer_0": "1451.6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=465, unit_price=800, discount_pct=0, region=EU, sku_total=1265 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1265", "v0": "372000", "v2": "37200", "v3": "372465"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 66 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "139", "answer_3": "111", "answer_1": "141", "answer_0": "126"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = $864; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $299. A fee of $162 is charged, then a deposit of $85 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "134", "answer_1": "119", "answer_3": "170", "answer_0": "141"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "32", "answer_2": "24", "answer_3": "8", "answer_1": "26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.385", "p1": "0.571", "p0": "0.429", "p2": "0.571"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1361 per year. It increases by 5% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1376.05", "answer_0": "1411.05", "answer_1": "1429.05", "answer_2": "1292.95"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "52", "age_1": "51", "age_3": "54", "age_2": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $633. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $8k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=137, unit_price=624, discount_pct=0, region=EU, sku_total=761 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "85625", "v1": "761", "v2": "8548", "v0": "85488"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $718. A fee of $194 is charged, then a deposit of $25 arrives. A second fee of $97 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "474", "answer_1": "478", "answer_0": "452", "answer_2": "425"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.538", "p1": "0.462", "p3": "0.500", "p2": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.364", "p1": "0.364", "p3": "0.600", "p0": "0.636"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 339:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $505. A fee of $131 is charged, then a deposit of $30 arrives. A second fee of $65 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "356", "answer_2": "315", "answer_3": "309", "answer_0": "339"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1966. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "57", "age_2": "59", "age_0": "58", "age_3": "56"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $447 per year. It increases by 5% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "458.35", "answer_2": "424.65", "answer_1": "469.35", "answer_3": "479.35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 21:15 to 22:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "0h30", "dur_1": "1h30", "dur_2": "1h00", "dur_0": "0h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = 2026-08-03; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Friday, March 13, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 61 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_3": "68", "answer_1": "78", "answer_2": "62"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = $918; priority = high.\n\nClaim to verify: \"the date is $918\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 5 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "21", "answer_0": "63", "answer_2": "14", "answer_3": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_3": "30", "answer_1": "35", "answer_2": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed March 02, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 12", "date_0": "March 10", "date_1": "March 08", "date_3": "March 10"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1316 per year. It increases by 15% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1513.4", "answer_0": "1489.4", "answer_2": "1118.6", "answer_3": "1481.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_1": "22", "answer_0": "64", "answer_3": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1983. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "44", "age_0": "43", "age_3": "41", "age_1": "42"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = 2026-10-15; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=399, unit_price=143, discount_pct=0, region=EU, sku_total=542 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "5705", "v0": "57057", "v3": "56658", "v1": "542"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.615", "p2": "0.615", "p0": "0.385", "p3": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = 2026-07-04; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Sunday, September 13, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 11:30 to 22:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "10h45", "dur_0": "10h00", "dur_2": "10h30", "dur_3": "9h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1771 per year. It increases by 5% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1809.55", "answer_1": "1859.55", "answer_3": "1821.55", "answer_2": "1682.45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 270 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "126", "answer_2": "113", "answer_0": "120", "answer_1": "105"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 722:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1601 per year. It increases by 25% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1200.75", "answer_0": "1951.25", "answer_1": "2001.25", "answer_3": "1987.25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $246 per year. It increases by 5% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "198.3", "answer_0": "213.3", "answer_2": "233.7", "answer_1": "258.3"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = ops; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = $893; priority = high.\n\nClaim to verify: \"the date is $893\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: owner = 2026-08-27; priority = low.\n\nClaim to verify: \"the owner is 2026-08-27\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $893. A fee of $68 is charged, then a deposit of $45 arrives. A second fee of $34 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "836", "answer_1": "820", "answer_3": "819", "answer_2": "857"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 11:15 to 12:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h45", "dur_1": "1h30", "dur_3": "0h30", "dur_0": "1h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = 2026-11-18; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.750", "p2": "0.750", "p3": "0.143"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 342 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "170", "answer_3": "164", "answer_2": "162", "answer_0": "168"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $505. A fee of $181 is charged, then a deposit of $70 arrives. A second fee of $90 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "280", "answer_0": "304", "answer_2": "280", "answer_1": "321"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 364 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "88", "answer_1": "86", "answer_0": "100", "answer_2": "117"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 11 people. A shopper bought a 33 gram box and a 330 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=745, unit_price=661, discount_pct=0, region=EU, sku_total=1406 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1406", "v2": "49244", "v3": "493190", "v0": "492445"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p0": "0.500", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $581. A fee of $77 is charged, then a deposit of $51 arrives. A second fee of $38 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "517", "answer_3": "491", "answer_1": "545", "answer_2": "506"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, June 03, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p0": "0.667", "p3": "0.167", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed January 10, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 21", "date_1": "January 20", "date_3": "January 22", "date_0": "January 23"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p2": "0.100", "p0": "0.150", "p3": "0.050"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 361 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "100", "answer_1": "105", "answer_2": "115", "answer_0": "96"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1978. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "50", "age_0": "48", "age_1": "47", "age_2": "49"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 118 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "63", "answer_2": "61", "answer_0": "45", "answer_3": "47"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1967. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_2": "59", "age_1": "57", "age_3": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.500", "p1": "0.833", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p0": "0.400", "p1": "0.600", "p2": "0.600"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 5 grams of salt per serving and serves 9 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = $250; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, June 19, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p2": "0.800", "p1": "0.800", "p3": "0.111"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed February 22, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 28", "date_2": "February 28", "date_0": "March 02", "date_3": "March 02"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed May 03, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 09", "date_2": "May 05", "date_1": "May 07", "date_0": "May 07"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $866. A fee of $48 is charged, then a deposit of $36 arrives. A second fee of $24 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "830", "answer_3": "811", "answer_2": "818", "answer_1": "812"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1231 per year. It increases by 20% this year, then a $49 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1411.2", "answer_1": "1477.2", "answer_0": "1428.2", "answer_2": "984.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.615", "p3": "0.583", "p2": "0.385", "p1": "0.385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $219 per year. It increases by 15% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "186.15", "answer_0": "222.85", "answer_3": "197.85", "answer_1": "251.85"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "15", "answer_0": "27", "answer_3": "18", "answer_2": "9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.467", "p1": "0.533", "p3": "0.429", "p2": "0.533"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 3 grams of salt per serving and serves 7 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = $214; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=587, unit_price=957, discount_pct=0, region=EU, sku_total=1544 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1544", "v3": "561172", "v0": "561759", "v2": "56175"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=996, unit_price=937, discount_pct=0, region=EU, sku_total=1933 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "932256", "v2": "93325", "v1": "1933", "v0": "933252"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 3 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 12 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=362, unit_price=383, discount_pct=0, region=EU, sku_total=745 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "745", "v2": "13864", "v3": "138284", "v0": "138646"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed March 15, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 20", "date_2": "March 22", "date_3": "March 22", "date_1": "March 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 335 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_2": "32", "answer_1": "30", "answer_3": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "8", "answer_2": "24", "answer_1": "26", "answer_0": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = 2026-09-12; priority = low.\n\nClaim to verify: \"the owner is 2026-09-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, January 28, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "13", "answer_3": "20", "answer_2": "8", "answer_0": "28"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 20:15 to 21:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "0h30", "dur_0": "0h00", "dur_1": "0h45", "dur_3": "-1h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=994, unit_price=985, discount_pct=0, region=EU, sku_total=1979 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "979090", "v3": "980084", "v2": "97909", "v1": "1979"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p1": "0.250", "p2": "0.083", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=557, unit_price=807, discount_pct=0, region=EU, sku_total=1364 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1364", "v0": "449499", "v2": "44949", "v3": "448942"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 297:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1963 per year. It increases by 15% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1668.55", "answer_1": "2257.45", "answer_0": "2235.45", "answer_3": "2195.45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "15", "answer_3": "6", "answer_0": "21", "answer_1": "17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = 2026-07-06; priority = high.\n\nClaim to verify: \"the status is 2026-07-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 268 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "96", "answer_2": "103", "answer_3": "104", "answer_1": "104"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.333", "p0": "0.667", "p3": "0.636"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1985. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "39", "age_0": "41", "age_2": "42", "age_1": "40"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.375", "p0": "0.250", "p2": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.600", "p2": "0.333", "p1": "0.333", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p2": "0.500", "p1": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1573 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1655.65", "answer_1": "1651.65", "answer_2": "1494.35", "answer_0": "1619.65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = 2026-07-25; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "14", "answer_3": "42", "answer_0": "56", "answer_1": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=204, unit_price=957, discount_pct=0, region=EU, sku_total=1161 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "195432", "v1": "1161", "v0": "195228", "v2": "19522"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Saturday, January 10, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "21", "answer_3": "9", "answer_2": "12", "answer_1": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1454 per year. It increases by 5% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1526.7", "answer_2": "1381.3", "answer_0": "1511.7", "answer_3": "1496.7"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = 2026-12-14; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1968. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_1": "57", "age_3": "60", "age_2": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 807:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = $671; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = 2026-11-25; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1959. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "67", "age_0": "65", "age_2": "66", "age_1": "64"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 304 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "112", "answer_0": "108", "answer_2": "111", "answer_3": "119"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1960. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "62", "age_2": "65", "age_0": "64", "age_1": "63"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, August 18, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 9 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 5 grams of salt per serving and serves 10 people. A shopper bought a 50 gram box and a 500 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.636", "p1": "0.636", "p0": "0.364", "p3": "0.300"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $748. A fee of $139 is charged, then a deposit of $62 arrives. A second fee of $69 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "606", "answer_1": "573", "answer_0": "602", "answer_3": "619"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = $731; priority = low.\n\nClaim to verify: \"the amount is $731\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 722:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1987. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "40", "age_0": "39", "age_3": "37", "age_1": "38"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 10 people. A shopper bought a 90 gram box and a 900 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=542, unit_price=870, discount_pct=0, region=EU, sku_total=1412 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "472082", "v2": "47154", "v1": "1412", "v0": "471540"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 570:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 309 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "79", "answer_0": "63", "answer_2": "66", "answer_3": "76"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 651:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = $125; priority = high.\n\nClaim to verify: \"the date is $125\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 357 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "90", "answer_3": "80", "answer_1": "75", "answer_2": "71"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = 2026-03-08; priority = high.\n\nClaim to verify: \"the date is 2026-03-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 108 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "27", "answer_1": "43", "answer_3": "41", "answer_0": "32"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.250", "p3": "0.714", "p0": "0.750"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1778 per year. It increases by 5% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1689.1", "answer_1": "1866.9", "answer_0": "1825.9", "answer_3": "1800.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, October 06, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 356 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "255", "answer_1": "278", "answer_0": "270", "answer_3": "276"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1973. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "50", "age_0": "51", "age_3": "53", "age_2": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = open; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 321 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "245", "answer_3": "256", "answer_2": "250", "answer_0": "252"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=770, unit_price=919, discount_pct=0, region=EU, sku_total=1689 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1689", "v2": "70763", "v3": "706860", "v0": "707630"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=519, unit_price=256, discount_pct=0, region=EU, sku_total=775 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "132864", "v1": "775", "v3": "133383", "v2": "13286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 457:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $242. A fee of $44 is charged, then a deposit of $48 arrives. A second fee of $22 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "230", "answer_1": "198", "answer_0": "224", "answer_3": "201"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed January 12, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 18", "date_2": "January 14", "date_0": "January 16", "date_1": "January 16"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = 2026-01-02; priority = high.\n\nClaim to verify: \"the status is 2026-01-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=507, unit_price=711, discount_pct=0, region=EU, sku_total=1218 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "36047", "v0": "360477", "v3": "360984", "v1": "1218"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_1": "25", "answer_0": "60", "answer_2": "18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = $24; priority = high.\n\nClaim to verify: \"the amount is $24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p3": "0.125", "p1": "0.500", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=454, unit_price=303, discount_pct=0, region=EU, sku_total=757 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "137108", "v0": "137562", "v1": "757", "v2": "13756"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=161, unit_price=352, discount_pct=0, region=EU, sku_total=513 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "513", "v2": "5667", "v3": "56833", "v0": "56672"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "16", "answer_2": "16", "answer_1": "20", "answer_0": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "21", "answer_2": "9", "answer_0": "30", "answer_1": "16"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 294 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "293", "answer_3": "280", "answer_1": "276", "answer_0": "288"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 977:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = 2026-02-26; priority = high.\n\nClaim to verify: \"the amount is 2026-02-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 682:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.333", "p3": "0.083", "p1": "0.417"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=659, unit_price=170, discount_pct=0, region=EU, sku_total=829 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "112030", "v3": "111371", "v2": "11203", "v1": "829"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.167", "p1": "0.500", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 745:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $826. A fee of $123 is charged, then a deposit of $28 arrives. A second fee of $61 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "670", "answer_2": "649", "answer_3": "644", "answer_1": "682"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, September 12, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 15:45 to 19:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "3h45", "dur_2": "3h45", "dur_0": "3h30", "dur_3": "2h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $27k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed January 03, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 14", "date_3": "January 15", "date_1": "January 13", "date_0": "January 16"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.500", "p3": "0.167", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=421, unit_price=165, discount_pct=0, region=EU, sku_total=586 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "69465", "v1": "586", "v2": "6946", "v3": "69044"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed March 14, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 23", "date_1": "March 20", "date_3": "March 22", "date_2": "March 25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, December 01, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.667", "p3": "0.167", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 10 people. A shopper bought a 80 gram box and a 800 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.600", "p3": "0.500", "p2": "0.400", "p1": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $30k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 558:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed October 22, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 04", "date_3": "October 31", "date_1": "October 29", "date_0": "November 02"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_0": "42", "answer_3": "12", "answer_1": "32"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=985, unit_price=966, discount_pct=0, region=EU, sku_total=1951 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "95151", "v3": "952495", "v0": "951510", "v1": "1951"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = $888; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p2": "0.417", "p3": "0.545", "p0": "0.583"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.167", "p0": "0.667", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.167", "p3": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "31", "answer_2": "24", "answer_3": "56", "answer_0": "80"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed October 12, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 22", "date_0": "October 26", "date_3": "October 24", "date_2": "October 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1984. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "41", "age_3": "38", "age_1": "39", "age_0": "40"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 619:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1974. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "51", "age_1": "50", "age_2": "52", "age_3": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 262:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=494, unit_price=542, discount_pct=0, region=EU, sku_total=1036 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "26774", "v1": "1036", "v0": "267748", "v3": "267254"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1966. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "60", "age_3": "57", "age_1": "58", "age_0": "59"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "10", "answer_1": "13", "answer_0": "25", "answer_3": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = $72; priority = high.\n\nClaim to verify: \"the status is $72\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 121 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "104", "answer_0": "90", "answer_1": "81", "answer_2": "109"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1987. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "39", "age_1": "38", "age_3": "41", "age_2": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $241. A fee of $49 is charged, then a deposit of $74 arrives. A second fee of $24 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "242", "answer_3": "260", "answer_2": "214", "answer_1": "257"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 9 grams of salt per serving and serves 6 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $345. A fee of $50 is charged, then a deposit of $55 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "353", "answer_1": "327", "answer_0": "325", "answer_2": "329"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 395 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "149", "answer_1": "157", "answer_3": "149", "answer_0": "168"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = $410; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1989. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "35", "age_3": "38", "age_2": "37", "age_0": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed June 02, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 09", "date_0": "June 11", "date_3": "June 11", "date_2": "June 13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1978. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "49", "age_1": "46", "age_0": "47", "age_2": "48"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = 2026-06-04; priority = low.\n\nClaim to verify: \"the owner is 2026-06-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.150", "p3": "0.050", "p0": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p3": "0.167", "p0": "0.500", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = 2026-03-11; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p3": "0.050", "p0": "0.100", "p1": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 12 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1970. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "56", "age_0": "54", "age_2": "55", "age_1": "53"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.467", "p1": "0.533", "p3": "0.429", "p2": "0.533"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.538", "p2": "0.462", "p3": "0.500", "p1": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 3 grams of salt per serving and serves 10 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, December 04, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = $932; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed July 03, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 11", "date_2": "July 13", "date_0": "July 15", "date_3": "July 13"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 4 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed September 04, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 12", "date_0": "September 14", "date_1": "September 10", "date_2": "September 16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 89 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "217", "answer_0": "225", "answer_1": "244", "answer_3": "237"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $653 per year. It increases by 5% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "645.65", "answer_2": "620.35", "answer_0": "656.65", "answer_1": "685.65"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = ops; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $660. A fee of $193 is charged, then a deposit of $28 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "405", "answer_1": "386", "answer_2": "375", "answer_0": "399"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $37k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = 2026-04-01; priority = low.\n\nClaim to verify: \"the amount is 2026-04-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, February 24, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "96", "answer_2": "48", "answer_3": "48", "answer_1": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 3 grams of salt per serving and serves 12 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p3": "0.050", "p2": "0.050", "p1": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $374. A fee of $62 is charged, then a deposit of $57 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "338", "answer_1": "348", "answer_3": "363", "answer_2": "317"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = $967; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 363 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "33", "answer_2": "30", "answer_3": "10", "answer_0": "24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $224. A fee of $91 is charged, then a deposit of $28 arrives. A second fee of $45 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "116", "answer_3": "122", "answer_2": "90", "answer_1": "120"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.500", "p2": "0.500", "p3": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $232. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "96", "answer_2": "40", "answer_1": "47", "answer_3": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "77", "answer_1": "35", "answer_2": "28", "answer_3": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $633. A fee of $101 is charged, then a deposit of $45 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "529", "answer_2": "503", "answer_1": "501", "answer_0": "527"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "30", "answer_0": "48", "answer_2": "18", "answer_1": "23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p1": "0.250", "p0": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 339 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "300", "answer_1": "292", "answer_0": "288", "answer_2": "300"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = $494; priority = low.\n\nClaim to verify: \"the status is $494\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, September 18, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=570, unit_price=608, discount_pct=0, region=EU, sku_total=1178 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1178", "v3": "347130", "v2": "34656", "v0": "346560"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.545", "p3": "0.500", "p1": "0.455", "p2": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.500", "p1": "0.833", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 139 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "34", "answer_1": "14", "answer_3": "42", "answer_0": "30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = 2026-10-07; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 301 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "340", "answer_0": "336", "answer_3": "321", "answer_1": "347"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $332. A fee of $50 is charged, then a deposit of $21 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "286", "answer_0": "278", "answer_3": "258", "answer_2": "290"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = $159; priority = high.\n\nClaim to verify: \"the amount is $159\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $7k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed February 18, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 23", "date_3": "February 25", "date_2": "February 27", "date_0": "February 25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 812:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 8 grams of salt per serving and serves 7 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 82 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "58", "answer_3": "52", "answer_0": "64", "answer_2": "66"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 675:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = $165; priority = high.\n\nClaim to verify: \"the owner is $165\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.417", "p0": "0.583", "p3": "0.545", "p1": "0.417"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Thursday, September 10, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 177 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "174", "answer_0": "192", "answer_3": "212", "answer_2": "189"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "14", "answer_0": "28", "answer_1": "16", "answer_3": "14"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_2": "35", "answer_1": "38", "answer_3": "21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed January 15, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 18", "date_0": "January 20", "date_3": "January 20", "date_2": "January 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.083", "p1": "0.333", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 224 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "80", "answer_0": "72", "answer_1": "63", "answer_3": "92"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 11:00 to 16:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "4h45", "dur_1": "5h30", "dur_3": "4h30", "dur_2": "5h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $636. A fee of $63 is charged, then a deposit of $51 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "605", "answer_0": "593", "answer_1": "580", "answer_3": "579"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $26k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=893, unit_price=383, discount_pct=0, region=EU, sku_total=1276 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1276", "v0": "342019", "v2": "34201", "v3": "342912"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=310, unit_price=616, discount_pct=0, region=EU, sku_total=926 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "190650", "v2": "19096", "v0": "190960", "v1": "926"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $539. A fee of $77 is charged, then a deposit of $48 arrives. A second fee of $38 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "457", "answer_2": "489", "answer_1": "494", "answer_0": "472"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 3 grams of salt per serving and serves 12 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.571", "p3": "0.333", "p0": "0.429", "p1": "0.571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $331. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $23k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_1": "33", "answer_0": "48", "answer_2": "30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed October 27, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 06", "date_1": "November 04", "date_0": "November 06", "date_2": "November 08"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 6 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=965, unit_price=922, discount_pct=0, region=EU, sku_total=1887 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "889730", "v1": "1887", "v2": "88973", "v3": "888765"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $670 per year. It increases by 15% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "770.5", "answer_0": "725.5", "answer_2": "569.5", "answer_3": "738.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = $450; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 3 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 2 grams of salt per serving and serves 12 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 300 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "104", "answer_0": "100", "answer_2": "92", "answer_3": "108"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 602:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 561:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 699:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 56 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "157", "answer_3": "127", "answer_0": "140", "answer_2": "142"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1595 per year. It increases by 20% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1897", "answer_3": "1884", "answer_1": "1914", "answer_2": "1276"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Thursday, October 22, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $220. A fee of $193 is charged, then a deposit of $42 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "3", "answer_0": "-27", "answer_3": "-2", "answer_1": "-12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.636", "p0": "0.364", "p3": "0.300", "p1": "0.636"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 4 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=763, unit_price=700, discount_pct=0, region=EU, sku_total=1463 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1463", "v2": "53410", "v0": "534100", "v3": "534863"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Sunday, September 13, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.385", "p2": "0.385", "p0": "0.615", "p3": "0.583"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1967. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "59", "age_1": "58", "age_2": "60", "age_3": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed March 10, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 18", "date_1": "March 16", "date_2": "March 20", "date_3": "March 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed February 10, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 15", "date_2": "February 15", "date_1": "February 13", "date_0": "February 13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $319 per year. It increases by 25% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "239.25", "answer_0": "372.75", "answer_3": "333.75", "answer_1": "398.75"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = $644; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 477:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = $655; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 381:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.200", "p3": "0.778", "p1": "0.200", "p0": "0.800"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_2": "42", "answer_0": "84", "answer_1": "48"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = closed; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p2": "0.250", "p3": "0.083", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 752:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $221. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.167", "p0": "0.667", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed July 21, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "August 04", "date_1": "July 31", "date_2": "August 02", "date_3": "August 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $903 per year. It increases by 25% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1096.75", "answer_1": "1128.75", "answer_0": "1108.75", "answer_2": "677.25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $879 per year. It increases by 15% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1010.85", "answer_0": "1000.85", "answer_2": "747.15", "answer_3": "1018.85"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: status = ops; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.333", "p2": "0.167", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed May 21, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 26", "date_0": "May 26", "date_2": "May 28", "date_1": "May 24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=104, unit_price=560, discount_pct=0, region=EU, sku_total=664 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "58136", "v1": "664", "v0": "58240", "v2": "5824"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.538", "p3": "0.417", "p0": "0.462", "p2": "0.538"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A recipe needs 8 grams of salt per serving and serves 10 people. A shopper bought a 80 gram box and a 800 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1406 per year. It increases by 25% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1716.5", "answer_1": "1757.5", "answer_2": "1054.5", "answer_3": "1750.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.150", "p3": "0.050", "p0": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = 2026-09-02; priority = high.\n\nClaim to verify: \"the status is 2026-09-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p2": "0.100", "p3": "0.050", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.167", "p3": "0.083", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = 2026-08-01; priority = high.\n\nClaim to verify: \"the date is 2026-08-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = closed; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.727", "p3": "0.200", "p0": "0.273", "p1": "0.727"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "15", "answer_0": "21", "answer_3": "6", "answer_1": "17"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $620. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 12 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = $391; priority = high.\n\nClaim to verify: \"the owner is $391\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.250", "p1": "0.500", "p0": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $666. A fee of $42 is charged, then a deposit of $72 arrives. A second fee of $21 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "660", "answer_3": "701", "answer_2": "680", "answer_0": "675"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1580 per year. It increases by 5% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1642", "answer_0": "1611", "answer_1": "1659", "answer_2": "1501"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1973. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "52", "age_3": "53", "age_1": "50", "age_0": "51"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.385", "p1": "0.385", "p0": "0.615", "p3": "0.583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = $61; priority = high.\n\nClaim to verify: \"the date is $61\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 442:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 11:15 to 17:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "5h00", "dur_3": "4h45", "dur_2": "5h15", "dur_1": "5h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "17", "answer_0": "25", "answer_3": "10", "answer_2": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1231 per year. It increases by 5% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1292.55", "answer_2": "1169.45", "answer_3": "1310.55", "answer_0": "1281.55"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $20k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $383. A fee of $124 is charged, then a deposit of $66 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "246", "answer_0": "263", "answer_2": "256", "answer_3": "290"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = pending; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = 2026-12-06; priority = high.\n\nClaim to verify: \"the date is 2026-12-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.625", "p1": "0.375", "p2": "0.375", "p3": "0.571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, October 19, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.714", "p2": "0.286", "p3": "0.667", "p1": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $13k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: date = $619; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, October 07, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 82 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "101", "answer_0": "81", "answer_1": "93", "answer_3": "99"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p1": "0.500", "p3": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_0": "56", "answer_2": "42", "answer_1": "44"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 13:45 to 20:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "5h30", "dur_2": "6h30", "dur_1": "6h30", "dur_0": "6h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = $515; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, December 18, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 265 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "101", "answer_3": "81", "answer_0": "90", "answer_2": "93"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 485:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1990. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "34", "age_2": "37", "age_1": "35", "age_0": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = $496; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = $705; priority = low.\n\nClaim to verify: \"the amount is $705\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: date = ops; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 116 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "373", "answer_3": "380", "answer_2": "372", "answer_0": "378"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = 2026-06-11; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $596. A fee of $120 is charged, then a deposit of $56 arrives. A second fee of $60 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "492", "answer_1": "447", "answer_0": "472", "answer_2": "491"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, October 05, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 106 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "79", "answer_2": "79", "answer_1": "64", "answer_0": "84"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.750", "p1": "0.222", "p2": "0.222", "p0": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=735, unit_price=130, discount_pct=0, region=EU, sku_total=865 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "94815", "v1": "865", "v2": "9555", "v0": "95550"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1986. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "38", "age_2": "39", "age_3": "40", "age_1": "37"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=955, unit_price=405, discount_pct=0, region=EU, sku_total=1360 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "386775", "v1": "1360", "v3": "385820", "v2": "38677"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = $671; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: amount = 2026-05-11; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: date = 2026-04-18; priority = low.\n\nClaim to verify: \"the date is 2026-04-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed April 23, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 28", "date_0": "April 30", "date_3": "April 30", "date_2": "April 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = 2026-06-13; priority = low.\n\nClaim to verify: \"the amount is 2026-06-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = 2026-06-16; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1997 per year. It increases by 20% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2396.4", "answer_0": "2358.4", "answer_2": "1597.6", "answer_3": "2353.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 243 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "149", "answer_3": "174", "answer_0": "160", "answer_2": "152"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p3": "0.125", "p2": "0.250", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = $869; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $478. A fee of $193 is charged, then a deposit of $44 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "233", "answer_2": "243", "answer_3": "263", "answer_1": "253"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = 2026-02-06; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = $275; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.467", "p2": "0.467", "p3": "0.500", "p0": "0.533"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=219, unit_price=599, discount_pct=0, region=EU, sku_total=818 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "130962", "v2": "13118", "v0": "131181", "v1": "818"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed January 09, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 17", "date_0": "January 15", "date_1": "January 13", "date_3": "January 15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.167", "p0": "0.667", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=594, unit_price=258, discount_pct=0, region=EU, sku_total=852 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "152658", "v2": "15325", "v0": "153252", "v1": "852"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $345. A fee of $141 is charged, then a deposit of $54 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "188", "answer_1": "200", "answer_2": "174", "answer_3": "195"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=852, unit_price=514, discount_pct=0, region=EU, sku_total=1366 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "437928", "v2": "43792", "v3": "437076", "v1": "1366"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = 2026-11-28; priority = high.\n\nClaim to verify: \"the amount is 2026-11-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: date = $113; priority = low.\n\nClaim to verify: \"the date is $113\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p1": "0.625", "p3": "0.125", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1990. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "35", "age_0": "36", "age_2": "37", "age_3": "34"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p1": "0.333", "p2": "0.333", "p3": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 5 grams of salt per serving and serves 3 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 846:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=466, unit_price=452, discount_pct=0, region=EU, sku_total=918 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "210632", "v3": "211098", "v1": "918", "v2": "21063"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p0": "0.500", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = open; priority = high.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.615", "p1": "0.615", "p3": "0.333", "p0": "0.385"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1112 per year. It increases by 25% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1360", "answer_1": "1390", "answer_3": "1324", "answer_2": "834"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $25k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 777:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "37", "answer_0": "49", "answer_2": "35", "answer_3": "14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 09:15 to 14:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "3h45", "dur_0": "4h30", "dur_1": "4h45", "dur_2": "5h00"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $34k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = 2026-04-24; priority = low.\n\nClaim to verify: \"the date is 2026-04-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 9 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, October 26, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p3": "0.250", "p1": "0.667", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "6", "answer_0": "12", "answer_1": "8", "answer_3": "6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.500", "p3": "0.167", "p1": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1966. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "60", "age_1": "59", "age_3": "58", "age_2": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "10", "answer_2": "15", "answer_0": "25", "answer_1": "17"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 152 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "37", "answer_0": "40", "answer_2": "42", "answer_3": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, June 09, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $40k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, September 07, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Saturday, November 14, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = $884; priority = low.\n\nClaim to verify: \"the status is $884\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.333", "p2": "0.167", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 262 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "135", "answer_3": "137", "answer_2": "153", "answer_1": "150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.250", "p1": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed September 15, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 25", "date_2": "September 27", "date_1": "September 23", "date_0": "September 25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 400:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 198 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "64", "answer_3": "44", "answer_1": "81", "answer_2": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 315 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "61", "answer_2": "58", "answer_0": "75", "answer_3": "95"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, February 18, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $238 per year. It increases by 25% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "178.5", "answer_3": "261.5", "answer_0": "254.5", "answer_1": "297.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, January 07, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.167", "p0": "0.333", "p3": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: owner = 2026-07-19; priority = high.\n\nClaim to verify: \"the owner is 2026-07-19\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed October 16, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 29", "date_0": "October 27", "date_1": "October 23", "date_3": "October 25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=823, unit_price=841, discount_pct=0, region=EU, sku_total=1664 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "691320", "v1": "1664", "v0": "692143", "v2": "69214"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1966. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "59", "age_2": "60", "age_1": "58", "age_3": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = 2026-06-24; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.125", "p2": "0.125", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = $234; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.700", "p2": "0.700", "p3": "0.222", "p0": "0.300"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 579:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 11:00 to 19:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "7h45", "dur_3": "7h00", "dur_0": "7h15", "dur_1": "8h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=706, unit_price=876, discount_pct=0, region=EU, sku_total=1582 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "61845", "v1": "1582", "v0": "618456", "v3": "617750"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, July 04, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 415:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1972. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "52", "age_3": "50", "age_1": "51", "age_2": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = $456; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 5 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 647:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 267:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = 2026-08-04; priority = high.\n\nClaim to verify: \"the amount is 2026-08-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "25", "answer_3": "30", "answer_1": "31", "answer_0": "55"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p2": "0.250", "p1": "0.500", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.385", "p1": "0.571", "p2": "0.571", "p0": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "15", "answer_1": "14", "answer_0": "24", "answer_2": "9"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Sunday, May 03, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 336:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1989. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "37", "age_2": "36", "age_0": "35", "age_1": "34"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 448:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = $725; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed October 04, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 10", "date_1": "October 08", "date_3": "October 10", "date_0": "October 08"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.167", "p3": "0.083", "p2": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, July 25, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $190. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=846, unit_price=493, discount_pct=0, region=EU, sku_total=1339 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "417078", "v3": "416232", "v2": "41707", "v1": "1339"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed July 12, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 26", "date_1": "July 22", "date_0": "July 24", "date_3": "July 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 179 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "153", "answer_3": "125", "answer_1": "157", "answer_0": "144"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 88 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "221", "answer_3": "205", "answer_0": "216", "answer_1": "196"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=194, unit_price=796, discount_pct=0, region=EU, sku_total=990 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "15442", "v0": "154424", "v1": "990", "v3": "154618"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=481, unit_price=652, discount_pct=0, region=EU, sku_total=1133 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1133", "v3": "314093", "v2": "31361", "v0": "313612"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, January 06, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Friday, June 05, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1987. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "38", "age_0": "39", "age_3": "37", "age_2": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 300:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = 2026-02-25; priority = high.\n\nClaim to verify: \"the status is 2026-02-25\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=222, unit_price=323, discount_pct=0, region=EU, sku_total=545 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "71484", "v1": "545", "v2": "7170", "v0": "71706"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "28", "answer_2": "16", "answer_0": "44", "answer_1": "23"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "48", "answer_1": "51", "answer_3": "24", "answer_0": "72"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 587:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1968. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "57", "age_2": "58", "age_1": "56", "age_3": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = 2026-05-23; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p3": "0.050", "p1": "0.200", "p0": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = 2026-03-03; priority = low.\n\nClaim to verify: \"the date is 2026-03-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed April 12, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 20", "date_1": "April 18", "date_2": "April 22", "date_3": "April 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1984. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "42", "age_3": "40", "age_1": "41", "age_2": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.800", "p1": "0.800", "p3": "0.111", "p0": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $423. A fee of $117 is charged, then a deposit of $69 arrives. A second fee of $58 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "289", "answer_1": "304", "answer_2": "296", "answer_0": "317"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.727", "p2": "0.273", "p1": "0.273", "p3": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, August 21, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=412, unit_price=790, discount_pct=0, region=EU, sku_total=1202 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1202", "v0": "325480", "v2": "32548", "v3": "325892"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 06:45 to 07:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h30", "dur_0": "0h00", "dur_1": "0h30", "dur_2": "0h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 990:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=255, unit_price=987, discount_pct=0, region=EU, sku_total=1242 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "25168", "v3": "251940", "v0": "251685", "v1": "1242"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, February 10, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $316. A fee of $149 is charged, then a deposit of $40 arrives. A second fee of $74 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "152", "answer_3": "112", "answer_0": "133", "answer_1": "157"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.400", "p1": "0.400", "p0": "0.600", "p3": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 583:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Sunday, April 19, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = $170; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 553:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p3": "0.429", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 74 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "58", "answer_3": "62", "answer_0": "48", "answer_1": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=470, unit_price=143, discount_pct=0, region=EU, sku_total=613 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "6721", "v3": "66740", "v1": "613", "v0": "67210"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_0": "28", "answer_1": "16", "answer_3": "16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=929, unit_price=261, discount_pct=0, region=EU, sku_total=1190 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "24246", "v1": "1190", "v0": "242469", "v3": "241540"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=560, unit_price=465, discount_pct=0, region=EU, sku_total=1025 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "260960", "v2": "26040", "v0": "260400", "v1": "1025"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1220 per year. It increases by 20% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1421", "answer_3": "1393", "answer_1": "1464", "answer_2": "976"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = 2026-06-23; priority = high.\n\nClaim to verify: \"the date is 2026-06-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 594:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.222", "p2": "0.778", "p1": "0.778"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = 2026-04-13; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.385", "p1": "0.571", "p0": "0.429", "p2": "0.571"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 245:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, December 19, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 14, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 24", "date_3": "April 26", "date_2": "April 26", "date_0": "April 28"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = $356; priority = high.\n\nClaim to verify: \"the date is $356\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 3 grams of salt per serving and serves 8 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p2": "0.050", "p3": "0.050", "p1": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = 2026-09-05; priority = high.\n\nClaim to verify: \"the amount is 2026-09-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_2": "9", "answer_1": "15", "answer_0": "27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, September 08, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1959. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "67", "age_0": "65", "age_1": "64", "age_2": "66"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 822:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1295 per year. It increases by 15% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1100.75", "answer_1": "1489.25", "answer_3": "1427.25", "answer_0": "1442.25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 122 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "112", "answer_1": "93", "answer_0": "96", "answer_2": "115"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1980. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "47", "age_2": "46", "age_1": "44", "age_0": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.250", "p2": "0.083", "p0": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $14k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1963. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "64", "age_1": "62", "age_0": "63", "age_3": "65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=568, unit_price=254, discount_pct=0, region=EU, sku_total=822 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "144840", "v1": "822", "v2": "14427", "v0": "144272"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 311:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $376. A fee of $171 is charged, then a deposit of $35 arrives. A second fee of $85 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "141", "answer_3": "167", "answer_2": "183", "answer_0": "155"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, September 19, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.600", "p1": "0.400", "p3": "0.500", "p2": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = $574; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "84", "answer_1": "48", "answer_2": "42", "answer_3": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $243. A fee of $97 is charged, then a deposit of $30 arrives. A second fee of $48 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "140", "answer_3": "126", "answer_0": "128", "answer_1": "132"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $606. A fee of $149 is charged, then a deposit of $38 arrives. A second fee of $74 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "412", "answer_0": "421", "answer_1": "400", "answer_3": "438"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = $619; priority = high.\n\nClaim to verify: \"the owner is $619\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 599:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, April 01, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Thursday, November 12, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $26k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 243:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Friday, September 04, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed August 10, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "August 15", "date_3": "August 15", "date_0": "August 13", "date_1": "August 13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1982. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "41", "age_2": "43", "age_3": "44", "age_0": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 12 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 282 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "179", "answer_1": "160", "answer_0": "168", "answer_2": "163"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.083", "p0": "0.167", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = 2026-12-08; priority = low.\n\nClaim to verify: \"the date is 2026-12-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $940 per year. It increases by 15% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1042", "answer_2": "799", "answer_1": "1081", "answer_3": "1081"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.333", "p1": "0.417", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 282:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "64", "answer_2": "24", "answer_1": "29", "answer_3": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = 2026-04-24; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "36", "answer_1": "30", "answer_2": "24", "answer_0": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed September 10, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 15", "date_1": "September 15", "date_0": "September 17", "date_3": "September 17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.533", "p2": "0.467", "p1": "0.467"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=310, unit_price=305, discount_pct=0, region=EU, sku_total=615 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "615", "v0": "94550", "v3": "94860", "v2": "9455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 166 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "36", "answer_3": "42", "answer_2": "64", "answer_0": "48"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = $478; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 7 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 241 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "104", "answer_2": "112", "answer_3": "104", "answer_0": "120"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 584:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, September 24, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.429", "p3": "0.538", "p1": "0.429", "p0": "0.571"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 2 grams of salt per serving and serves 6 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.636", "p2": "0.636", "p3": "0.300", "p0": "0.364"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed August 04, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 08", "date_3": "August 10", "date_0": "August 10", "date_2": "August 08"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1987. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "40", "age_2": "39", "age_0": "38", "age_1": "37"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 122 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_3": "45", "answer_1": "38", "answer_2": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 8 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 532:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1980. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "45", "age_3": "47", "age_2": "46", "age_1": "44"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1715 per year. It increases by 25% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2127.75", "answer_3": "2102.75", "answer_2": "1286.25", "answer_1": "2143.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 124 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_3": "52", "answer_1": "77", "answer_2": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 09:15 to 14:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "5h00", "dur_2": "5h00", "dur_0": "4h30", "dur_3": "4h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.273", "p2": "0.273", "p3": "0.700", "p0": "0.727"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $36k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 825:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = 2026-09-23; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 390 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "43", "answer_2": "32", "answer_0": "45", "answer_1": "53"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "34", "answer_0": "70", "answer_3": "42", "answer_2": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=582, unit_price=410, discount_pct=0, region=EU, sku_total=992 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "992", "v3": "238038", "v2": "23862", "v0": "238620"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed December 02, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 12", "date_3": "December 14", "date_0": "December 16", "date_2": "December 18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1672 per year. It increases by 20% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1968.4", "answer_2": "1337.6", "answer_3": "1957.4", "answer_1": "2006.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, January 06, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 09:45 to 17:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "8h00", "dur_0": "7h30", "dur_3": "7h00", "dur_2": "8h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, October 27, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 396 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "270", "answer_3": "254", "answer_1": "267", "answer_2": "254"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 538:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.600", "p1": "0.600", "p0": "0.400", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1965. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "60", "age_3": "59", "age_0": "61", "age_2": "62"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=245, unit_price=775, discount_pct=0, region=EU, sku_total=1020 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1020", "v2": "18987", "v3": "190120", "v0": "189875"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=501, unit_price=709, discount_pct=0, region=EU, sku_total=1210 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "355209", "v3": "354708", "v2": "35520", "v1": "1210"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=508, unit_price=364, discount_pct=0, region=EU, sku_total=872 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "184404", "v2": "18491", "v1": "872", "v0": "184912"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed January 22, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 01", "date_0": "February 03", "date_3": "February 01", "date_1": "January 30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p0": "0.333", "p3": "0.083", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $550. A fee of $80 is charged, then a deposit of $86 arrives. A second fee of $40 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "529", "answer_2": "490", "answer_0": "516", "answer_3": "527"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=589, unit_price=846, discount_pct=0, region=EU, sku_total=1435 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "497705", "v2": "49829", "v0": "498294", "v1": "1435"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed January 28, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 04", "date_2": "February 02", "date_3": "February 04", "date_1": "February 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 11 people. A shopper bought a 66 gram box and a 660 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 9 grams of salt per serving and serves 3 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 7 grams of salt per serving and serves 6 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 346:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1977. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "48", "age_3": "47", "age_0": "49", "age_2": "50"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = 2026-06-17; priority = low.\n\nClaim to verify: \"the date is 2026-06-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 5 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed October 22, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 05", "date_2": "November 03", "date_3": "November 03", "date_1": "November 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.150", "p0": "0.200", "p1": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, January 23, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1971. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "55", "age_3": "57", "age_2": "56", "age_1": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.444", "p0": "0.500", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 395:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1973. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "53", "age_3": "51", "age_1": "52", "age_2": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 4 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1978. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "48", "age_2": "49", "age_1": "47", "age_3": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $303 per year. It increases by 10% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "272.7", "answer_1": "333.3", "answer_0": "298.3", "answer_3": "338.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.500", "p3": "0.444", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Thursday, December 24, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $266. A fee of $106 is charged, then a deposit of $21 arrives. A second fee of $53 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "115", "answer_2": "152", "answer_3": "122", "answer_0": "128"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=773, unit_price=450, discount_pct=0, region=EU, sku_total=1223 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "348623", "v1": "1223", "v2": "34785", "v0": "347850"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 716:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1986. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "37", "age_2": "40", "age_0": "39", "age_1": "38"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.625", "p2": "0.333", "p0": "0.667", "p1": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.143", "p1": "0.750", "p2": "0.750", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $32k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $558. A fee of $57 is charged, then a deposit of $81 arrives. A second fee of $28 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "583", "answer_0": "554", "answer_1": "579", "answer_3": "558"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = 2026-02-22; priority = low.\n\nClaim to verify: \"the status is 2026-02-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 97 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "124", "answer_3": "93", "answer_2": "100", "answer_0": "108"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 10:00 to 22:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "12h30", "dur_0": "11h45", "dur_2": "12h15", "dur_3": "11h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 12:00 to 19:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "6h45", "dur_0": "6h30", "dur_3": "6h15", "dur_1": "7h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p3": "0.125", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed May 25, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 28", "date_0": "May 28", "date_2": "May 30", "date_3": "May 30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_1": "18", "answer_3": "9", "answer_2": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = 2026-05-17; priority = low.\n\nClaim to verify: \"the amount is 2026-05-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, August 08, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1978. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "47", "age_0": "48", "age_2": "49", "age_3": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = closed; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p3": "0.050", "p1": "0.150", "p2": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 708:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "6", "answer_0": "24", "answer_1": "20", "answer_2": "18"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $602. A fee of $191 is charged, then a deposit of $27 arrives. A second fee of $95 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "332", "answer_2": "327", "answer_0": "343", "answer_1": "340"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = $417; priority = low.\n\nClaim to verify: \"the amount is $417\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 07:15 to 19:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "11h30", "dur_1": "12h00", "dur_2": "11h45", "dur_3": "11h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.200", "p2": "0.727", "p0": "0.273", "p1": "0.727"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.556", "p3": "0.500", "p2": "0.444", "p1": "0.444"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 192 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "138", "answer_1": "146", "answer_0": "144", "answer_3": "163"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = $360; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p1": "0.250", "p0": "0.200", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=669, unit_price=145, discount_pct=0, region=EU, sku_total=814 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "97005", "v3": "96336", "v1": "814", "v2": "9700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 09:00 to 16:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "6h15", "dur_1": "7h15", "dur_2": "7h00", "dur_0": "6h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.444", "p0": "0.500", "p1": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1963. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "62", "age_3": "65", "age_2": "64", "age_0": "63"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.222", "p3": "0.750", "p2": "0.222", "p0": "0.778"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $860. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 544:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 17:30 to 17:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "0h15", "dur_0": "-1h45", "dur_1": "0h15", "dur_3": "-1h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $737. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=548, unit_price=348, discount_pct=0, region=EU, sku_total=896 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "190704", "v3": "190156", "v2": "19070", "v1": "896"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1958. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "67", "age_3": "64", "age_0": "66", "age_1": "65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = 2026-12-08; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed April 25, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 06", "date_1": "May 04", "date_2": "May 05", "date_0": "May 07"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1976. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "47", "age_2": "49", "age_3": "46", "age_0": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 259 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "100", "answer_1": "96", "answer_2": "72", "answer_0": "81"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=946, unit_price=661, discount_pct=0, region=EU, sku_total=1607 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "624360", "v0": "625306", "v2": "62530", "v1": "1607"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 205 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "427", "answer_2": "440", "answer_1": "414", "answer_0": "432"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = 2026-05-10; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p2": "0.100", "p3": "0.050", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $215. A fee of $56 is charged, then a deposit of $29 arrives. A second fee of $28 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "137", "answer_0": "160", "answer_2": "181", "answer_3": "157"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $542. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = $154; priority = low.\n\nClaim to verify: \"the owner is $154\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = pending; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1911 per year. It increases by 5% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1937.55", "answer_1": "2006.55", "answer_0": "1977.55", "answer_2": "1815.45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p3": "0.125", "p1": "0.375", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 9 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=946, unit_price=931, discount_pct=0, region=EU, sku_total=1877 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "880726", "v1": "1877", "v2": "88072", "v3": "881672"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.273", "p1": "0.273", "p0": "0.727", "p3": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.667", "p3": "0.625", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_2": "15", "answer_1": "19", "answer_0": "27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 7 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "14", "answer_0": "30", "answer_3": "20", "answer_2": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 07:30 to 15:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "7h00", "dur_1": "7h45", "dur_2": "7h15", "dur_3": "6h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $374. A fee of $50 is charged, then a deposit of $27 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "356", "answer_0": "326", "answer_3": "348", "answer_2": "296"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 12 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = $648; priority = high.\n\nClaim to verify: \"the amount is $648\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: status = $353; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.500", "p2": "0.500", "p3": "0.444"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed October 23, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 02", "date_2": "November 02", "date_0": "November 04", "date_1": "October 31"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $813 per year. It increases by 15% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "934.95", "answer_2": "691.05", "answer_3": "946.95", "answer_0": "922.95"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p2": "0.667", "p1": "0.667", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 02, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 05", "date_3": "April 07", "date_1": "April 05", "date_0": "April 07"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p3": "0.125", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Sunday, December 06, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.375", "p2": "0.125", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.375", "p2": "0.556", "p1": "0.556", "p0": "0.444"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = 2026-01-25; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, May 28, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.500", "p0": "0.333", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.500", "p3": "0.167", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 15:45 to 21:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "4h30", "dur_2": "4h45", "dur_3": "4h15", "dur_1": "5h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 17:45 to 18:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "0h30", "dur_1": "1h00", "dur_0": "0h15", "dur_3": "0h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.333", "p1": "0.417", "p3": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 176 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "39", "answer_0": "36", "answer_1": "18", "answer_3": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $254 per year. It increases by 10% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "279.4", "answer_0": "266.4", "answer_3": "247.4", "answer_2": "228.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 343 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "162", "answer_3": "147", "answer_1": "145", "answer_2": "157"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Thursday, September 03, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed November 14, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 25", "date_0": "November 27", "date_1": "November 24", "date_3": "November 26"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p3": "0.167", "p0": "0.667", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1361 per year. It increases by 10% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1224.9", "answer_1": "1497.1", "answer_0": "1476.1", "answer_3": "1451.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 07:45 to 20:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "11h45", "dur_0": "11h30", "dur_1": "12h15", "dur_3": "11h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = 2026-08-16; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.050", "p0": "0.200", "p2": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "21", "answer_3": "35", "answer_1": "26", "answer_0": "56"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p0": "0.400", "p2": "0.600", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 219 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "265", "answer_1": "259", "answer_2": "257", "answer_0": "252"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $486. A fee of $176 is charged, then a deposit of $54 arrives. A second fee of $88 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "289", "answer_3": "290", "answer_0": "276", "answer_2": "258"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p0": "0.500", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = $560; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_1": "30", "answer_0": "48", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 381 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "42", "answer_3": "45", "answer_2": "57", "answer_1": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1975. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "50", "age_0": "49", "age_1": "48", "age_3": "51"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 325 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "18", "answer_3": "47", "answer_0": "30", "answer_2": "42"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 311 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "80", "answer_3": "42", "answer_0": "60", "answer_1": "76"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=490, unit_price=771, discount_pct=0, region=EU, sku_total=1261 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "37779", "v3": "377300", "v0": "377790", "v1": "1261"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $240. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p3": "0.083", "p1": "0.250", "p0": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p0": "0.150", "p3": "0.050", "p2": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 564:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=150, unit_price=578, discount_pct=0, region=EU, sku_total=728 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "8670", "v3": "86550", "v0": "86700", "v1": "728"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.800", "p3": "0.111", "p1": "0.800", "p0": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed November 20, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 23", "date_0": "November 25", "date_3": "November 25", "date_2": "November 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.273", "p1": "0.727", "p3": "0.200", "p2": "0.727"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = $152; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "16", "answer_2": "16", "answer_1": "20", "answer_0": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed October 07, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 15", "date_2": "October 17", "date_3": "October 17", "date_0": "October 19"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed August 06, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "August 22", "date_0": "August 20", "date_3": "August 18", "date_1": "August 16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1982. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "43", "age_3": "42", "age_0": "44", "age_2": "45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $503 per year. It increases by 20% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "603.6", "answer_3": "582.6", "answer_2": "402.4", "answer_0": "559.6"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.286", "p3": "0.667", "p0": "0.714", "p2": "0.286"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.167", "p3": "0.167", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.333", "p1": "0.667", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1874 per year. It increases by 20% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2250.8", "answer_2": "1499.2", "answer_0": "2238.8", "answer_1": "2248.8"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 266 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "139", "answer_3": "113", "answer_1": "106", "answer_0": "126"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 201 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_3": "27", "answer_2": "30", "answer_1": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "36", "answer_3": "18", "answer_0": "54", "answer_1": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "10", "answer_1": "12", "answer_2": "10", "answer_0": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=208, unit_price=286, discount_pct=0, region=EU, sku_total=494 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "5948", "v1": "494", "v0": "59488", "v3": "59280"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Sunday, September 13, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $370 per year. It increases by 5% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "388.5", "answer_2": "351.5", "answer_0": "357.5", "answer_3": "390.5"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 143 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "93", "answer_1": "86", "answer_0": "96", "answer_3": "89"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 5 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $309. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 10 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 323 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "281", "answer_0": "288", "answer_1": "297", "answer_3": "291"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $740. A fee of $172 is charged, then a deposit of $34 arrives. A second fee of $86 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "543", "answer_1": "491", "answer_0": "516", "answer_2": "503"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p3": "0.125", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1968 per year. It increases by 10% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2165.8", "answer_1": "2164.8", "answer_0": "2140.8", "answer_2": "1771.2"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1746 per year. It increases by 15% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2007.9", "answer_2": "1484.1", "answer_3": "1969.9", "answer_0": "1979.9"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = $297; priority = high.\n\nClaim to verify: \"the date is $297\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=505, unit_price=935, discount_pct=0, region=EU, sku_total=1440 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "472175", "v2": "47217", "v3": "472680", "v1": "1440"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 215 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "280", "answer_1": "266", "answer_3": "260", "answer_2": "272"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1978. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "46", "age_1": "45", "age_3": "44", "age_2": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 71 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "33", "answer_0": "27", "answer_1": "40", "answer_3": "12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1984. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "40", "age_2": "41", "age_3": "42", "age_1": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $826 per year. It increases by 25% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "619.5", "answer_1": "1032.5", "answer_0": "1019.5", "answer_3": "1034.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = $289; priority = low.\n\nClaim to verify: \"the owner is $289\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "25", "answer_0": "40", "answer_2": "15", "answer_1": "20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Wednesday, March 04, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=119, unit_price=265, discount_pct=0, region=EU, sku_total=384 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "31535", "v1": "384", "v2": "3153", "v3": "31654"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.150", "p1": "0.200", "p2": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $443. A fee of $65 is charged, then a deposit of $80 arrives. A second fee of $32 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "413", "answer_0": "426", "answer_1": "397", "answer_2": "415"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p3": "0.400", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.083", "p2": "0.250", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=547, unit_price=889, discount_pct=0, region=EU, sku_total=1436 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "485736", "v1": "1436", "v0": "486283", "v2": "48628"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Sunday, January 25, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1981. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "42", "age_0": "44", "age_1": "43", "age_2": "45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 273 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "238", "answer_1": "262", "answer_0": "252", "answer_3": "242"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = 2026-06-25; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=745, unit_price=216, discount_pct=0, region=EU, sku_total=961 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "161665", "v2": "16092", "v1": "961", "v0": "160920"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 14:00 to 20:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "6h15", "dur_3": "5h30", "dur_0": "6h00", "dur_1": "6h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 987:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.467", "p2": "0.500", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p2": "0.125", "p0": "0.250", "p3": "0.125"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.167", "p2": "0.167", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 327 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "240", "answer_3": "251", "answer_2": "234", "answer_1": "232"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_2": "8", "answer_1": "15", "answer_3": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 3 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = 2026-05-17; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1972. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "54", "age_3": "56", "age_1": "53", "age_2": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed November 17, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 23", "date_3": "November 25", "date_0": "November 25", "date_2": "November 23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.462", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed May 09, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 19", "date_2": "May 18", "date_0": "May 20", "date_1": "May 17"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $519. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 358 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "26", "answer_2": "24", "answer_1": "44", "answer_0": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, April 16, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $956 per year. It increases by 20% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1147.2", "answer_2": "764.8", "answer_0": "1118.2", "answer_3": "1087.2"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=992, unit_price=243, discount_pct=0, region=EU, sku_total=1235 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1235", "v0": "241056", "v2": "24105", "v3": "242048"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 8 grams of salt per serving and serves 12 people. A shopper bought a 96 gram box and a 960 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $349. A fee of $159 is charged, then a deposit of $85 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "196", "answer_3": "179", "answer_1": "166", "answer_2": "216"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p2": "0.050", "p3": "0.050", "p1": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = 2026-07-10; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p1": "0.250", "p3": "0.050", "p0": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=997, unit_price=223, discount_pct=0, region=EU, sku_total=1220 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1220", "v2": "22233", "v3": "221334", "v0": "222331"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1068 per year. It increases by 15% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1208.2", "answer_1": "1228.2", "answer_2": "907.8", "answer_3": "1237.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = $276; priority = high.\n\nClaim to verify: \"the date is $276\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 3 grams of salt per serving and serves 7 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1242 per year. It increases by 15% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1428.3", "answer_2": "1055.7", "answer_0": "1380.3", "answer_3": "1365.3"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 09:30 to 09:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h15", "dur_1": "0h15", "dur_0": "-1h45", "dur_2": "0h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1987. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "36", "age_3": "35", "age_0": "37", "age_2": "38"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.385", "p0": "0.615", "p2": "0.385", "p3": "0.583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 10:45 to 18:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "8h00", "dur_2": "7h45", "dur_3": "7h00", "dur_0": "7h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.200", "p1": "0.200", "p0": "0.800", "p3": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed August 28, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 06", "date_0": "September 04", "date_1": "September 02", "date_3": "September 04"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = 2026-09-25; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 308:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 10:30 to 12:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "0h30", "dur_0": "1h00", "dur_2": "1h15", "dur_1": "1h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = 2026-05-12; priority = low.\n\nClaim to verify: \"the amount is 2026-05-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = 2026-02-11; priority = high.\n\nClaim to verify: \"the date is 2026-02-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 324 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "111", "answer_3": "138", "answer_0": "125", "answer_2": "105"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 7 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1971. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "55", "age_2": "56", "age_1": "54", "age_3": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $32k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 08:45 to 14:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "4h30", "dur_1": "5h30", "dur_2": "5h30", "dur_0": "5h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = $95; priority = high.\n\nClaim to verify: \"the amount is $95\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: status = $211; priority = high.\n\nClaim to verify: \"the status is $211\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = $748; priority = high.\n\nClaim to verify: \"the amount is $748\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $423. A fee of $73 is charged, then a deposit of $65 arrives. A second fee of $36 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "371", "answer_2": "384", "answer_3": "382", "answer_0": "379"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $541. A fee of $105 is charged, then a deposit of $65 arrives. A second fee of $52 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "455", "answer_2": "477", "answer_1": "427", "answer_0": "449"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = 2026-12-26; priority = low.\n\nClaim to verify: \"the amount is 2026-12-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 297 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "111", "answer_1": "94", "answer_3": "98", "answer_0": "96"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = pending; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = 2026-09-17; priority = low.\n\nClaim to verify: \"the date is 2026-09-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Thursday, November 05, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $347. A fee of $113 is charged, then a deposit of $71 arrives. A second fee of $56 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "249", "answer_3": "269", "answer_2": "224", "answer_1": "224"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.286", "p0": "0.714", "p3": "0.667", "p2": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Sunday, May 03, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.125", "p2": "0.375", "p1": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = 2026-08-06; priority = high.\n\nClaim to verify: \"the date is 2026-08-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "64", "answer_3": "16", "answer_1": "50", "answer_2": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1965. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "61", "age_1": "60", "age_2": "62", "age_3": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p3": "0.125", "p2": "0.250", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = $697; priority = high.\n\nClaim to verify: \"the amount is $697\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Thursday, March 26, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p2": "0.571", "p0": "0.429", "p3": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 18:30 to 21:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h45", "dur_0": "2h15", "dur_2": "2h30", "dur_3": "1h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.083", "p0": "0.250", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 26, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 04", "date_0": "May 04", "date_1": "May 02", "date_2": "May 02"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.333", "p0": "0.250", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = 2026-07-10; priority = low.\n\nClaim to verify: \"the status is 2026-07-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 276 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "67", "answer_2": "70", "answer_1": "78", "answer_0": "72"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=368, unit_price=847, discount_pct=0, region=EU, sku_total=1215 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "31169", "v0": "311696", "v1": "1215", "v3": "311328"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, August 12, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed October 09, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 16", "date_0": "October 20", "date_2": "October 22", "date_3": "October 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.273", "p3": "0.700", "p1": "0.273", "p0": "0.727"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 278 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_1": "114", "answer_3": "108", "answer_2": "128"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $646. A fee of $131 is charged, then a deposit of $44 arrives. A second fee of $65 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "517", "answer_2": "478", "answer_0": "494", "answer_3": "487"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $26k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1006 per year. It increases by 20% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "804.8", "answer_3": "1163.2", "answer_0": "1170.2", "answer_1": "1207.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Thursday, April 16, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = $54; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, April 18, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.467", "p0": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed October 14, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 21", "date_2": "October 19", "date_0": "October 21", "date_1": "October 19"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, December 06, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "18", "answer_2": "9", "answer_1": "12", "answer_3": "9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 285 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "93", "answer_0": "96", "answer_3": "103", "answer_1": "115"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1980. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "48", "age_2": "47", "age_0": "46", "age_1": "45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed October 04, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 15", "date_1": "October 11", "date_3": "October 13", "date_0": "October 13"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 6 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 261 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "426", "answer_0": "432", "answer_1": "438", "answer_3": "417"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $345. A fee of $166 is charged, then a deposit of $88 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "172", "answer_2": "197", "answer_0": "184", "answer_3": "177"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1990. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "33", "age_3": "36", "age_2": "35", "age_0": "34"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.571", "p1": "0.571", "p3": "0.333", "p0": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed March 07, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 14", "date_0": "March 17", "date_3": "March 16", "date_2": "March 19"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 358 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "19", "answer_3": "34", "answer_1": "9", "answer_0": "27"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed February 18, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 24", "date_2": "February 26", "date_0": "February 24", "date_1": "February 22"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $439 per year. It increases by 5% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "448.95", "answer_1": "460.95", "answer_2": "417.05", "answer_3": "466.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed June 23, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 28", "date_1": "June 26", "date_0": "June 26", "date_2": "June 28"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "20", "answer_3": "8", "answer_0": "28", "answer_1": "22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $497. A fee of $144 is charged, then a deposit of $58 arrives. A second fee of $72 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "319", "answer_3": "349", "answer_0": "339", "answer_1": "363"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = $54; priority = high.\n\nClaim to verify: \"the date is $54\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p1": "0.150", "p2": "0.050", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=242, unit_price=665, discount_pct=0, region=EU, sku_total=907 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "16093", "v3": "160688", "v0": "160930", "v1": "907"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 2 grams of salt per serving and serves 7 people. A shopper bought a 14 gram box and a 140 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $7k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1232 per year. It increases by 25% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1521", "answer_0": "1502", "answer_2": "924", "answer_1": "1540"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=431, unit_price=286, discount_pct=0, region=EU, sku_total=717 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "717", "v0": "123266", "v2": "12326", "v3": "123697"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed October 14, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 22", "date_0": "October 26", "date_2": "October 28", "date_3": "October 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=692, unit_price=880, discount_pct=0, region=EU, sku_total=1572 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "60896", "v1": "1572", "v0": "608960", "v3": "609652"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 9 grams of salt per serving and serves 11 people. A shopper bought a 99 gram box and a 990 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p2": "0.050", "p1": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 730:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 276:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=477, unit_price=263, discount_pct=0, region=EU, sku_total=740 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "12545", "v3": "124974", "v0": "125451", "v1": "740"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 308:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $955 per year. It increases by 20% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1136", "answer_2": "764", "answer_1": "1146", "answer_3": "1110"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $897. A fee of $136 is charged, then a deposit of $73 arrives. A second fee of $68 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "757", "answer_1": "748", "answer_0": "766", "answer_3": "763"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $997 per year. It increases by 5% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1020.85", "answer_1": "1046.85", "answer_2": "947.15", "answer_3": "1032.85"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.150", "p0": "0.100", "p2": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Form fields: date = $557; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.375", "p2": "0.125", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p0": "0.333", "p1": "0.667", "p3": "0.273"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = 2026-03-06; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $399 per year. It increases by 5% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "373.95", "answer_1": "418.95", "answer_3": "413.95", "answer_2": "379.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 122 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "65", "answer_0": "60", "answer_3": "63", "answer_2": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.050", "p0": "0.200", "p2": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $675. A fee of $113 is charged, then a deposit of $50 arrives. A second fee of $56 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "586", "answer_2": "559", "answer_3": "560", "answer_0": "556"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 208:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "24", "answer_3": "18", "answer_0": "36", "answer_2": "18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = 2026-07-13; priority = high.\n\nClaim to verify: \"the amount is 2026-07-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_3": "15", "answer_0": "27", "answer_1": "17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=351, unit_price=207, discount_pct=0, region=EU, sku_total=558 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "7265", "v3": "73008", "v0": "72657", "v1": "558"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $580. A fee of $170 is charged, then a deposit of $49 arrives. A second fee of $85 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "369", "answer_3": "400", "answer_1": "361", "answer_0": "374"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1982. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "44", "age_2": "43", "age_0": "42", "age_1": "41"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.250", "p1": "0.500", "p0": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 3 grams of salt per serving and serves 8 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, August 09, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.667", "p2": "0.500", "p1": "0.833"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.250", "p2": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1969. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "56", "age_3": "54", "age_2": "57", "age_1": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $266. A fee of $63 is charged, then a deposit of $64 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "220", "answer_1": "226", "answer_3": "212", "answer_0": "236"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $225. A fee of $174 is charged, then a deposit of $53 arrives. A second fee of $87 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "44", "answer_0": "17", "answer_2": "42", "answer_3": "22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.538", "p2": "0.462", "p1": "0.462", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "15", "answer_2": "15", "answer_1": "20", "answer_0": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1965. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "62", "age_3": "63", "age_1": "60", "age_0": "61"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $757. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 252 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "191", "answer_1": "220", "answer_0": "200", "answer_2": "208"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $35k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 6 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p1": "0.571", "p0": "0.429", "p2": "0.571"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $770. A fee of $138 is charged, then a deposit of $41 arrives. A second fee of $69 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "625", "answer_2": "597", "answer_3": "616", "answer_0": "604"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 742:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.500", "p3": "0.125", "p0": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1960. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "64", "age_0": "65", "age_3": "67", "age_2": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1986. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "39", "age_1": "38", "age_3": "41", "age_2": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 11:00 to 17:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "5h30", "dur_0": "6h15", "dur_2": "6h30", "dur_1": "6h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $279 per year. It increases by 15% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "237.15", "answer_3": "314.85", "answer_1": "320.85", "answer_0": "286.85"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $781. A fee of $168 is charged, then a deposit of $38 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "567", "answer_2": "580", "answer_3": "592", "answer_1": "588"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.462", "p0": "0.538", "p2": "0.462"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $829 per year. It increases by 10% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "746.1", "answer_0": "885.9", "answer_3": "866.9", "answer_1": "911.9"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed July 06, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 18", "date_1": "July 16", "date_0": "July 20", "date_2": "July 18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, October 02, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed December 15, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 30", "date_0": "December 28", "date_3": "December 26", "date_1": "December 24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p0": "0.500", "p3": "0.462"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = $877; priority = high.\n\nClaim to verify: \"the amount is $877\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = $11; priority = high.\n\nClaim to verify: \"the date is $11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1965. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "62", "age_3": "59", "age_1": "60", "age_0": "61"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.429", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed April 07, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 15", "date_1": "April 15", "date_0": "April 17", "date_3": "April 17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 933:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.467", "p1": "0.500", "p0": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $781. A fee of $89 is charged, then a deposit of $34 arrives. A second fee of $44 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "710", "answer_0": "682", "answer_2": "687", "answer_1": "655"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $414. A fee of $144 is charged, then a deposit of $21 arrives. A second fee of $72 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "203", "answer_0": "219", "answer_3": "236", "answer_1": "209"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.462", "p0": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p3": "0.400", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 288 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "142", "answer_3": "145", "answer_1": "156", "answer_0": "160"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed July 04, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 16", "date_0": "July 14", "date_1": "July 11", "date_3": "July 13"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = ops; priority = low.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 12 people. A shopper bought a 96 gram box and a 960 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=770, unit_price=724, discount_pct=0, region=EU, sku_total=1494 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "55748", "v0": "557480", "v3": "558250", "v1": "1494"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed May 04, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 11", "date_1": "May 09", "date_2": "May 13", "date_0": "May 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "48", "answer_1": "22", "answer_2": "16", "answer_0": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = $924; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 304 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "98", "answer_1": "95", "answer_3": "96", "answer_2": "110"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed January 01, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 11", "date_3": "January 11", "date_1": "January 09", "date_0": "January 13"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed January 17, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 26", "date_2": "January 25", "date_0": "January 27", "date_1": "January 24"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p1": "0.250", "p3": "0.083", "p0": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed April 19, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 20", "date_3": "April 24", "date_0": "April 22", "date_1": "April 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=372, unit_price=137, discount_pct=0, region=EU, sku_total=509 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "509", "v3": "51336", "v2": "5096", "v0": "50964"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 15:30 to 21:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "5h30", "dur_2": "6h00", "dur_1": "6h00", "dur_3": "5h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_3": "24", "answer_1": "28", "answer_2": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 08:45 to 12:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h00", "dur_1": "3h30", "dur_0": "2h45", "dur_3": "2h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1124 per year. It increases by 5% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1067.8", "answer_1": "1180.2", "answer_0": "1147.2", "answer_3": "1125.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $401 per year. It increases by 25% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "489.25", "answer_1": "501.25", "answer_2": "300.75", "answer_3": "478.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "28", "answer_1": "16", "answer_2": "12", "answer_3": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 9 grams of salt per serving and serves 3 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $628. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = $839; priority = low.\n\nClaim to verify: \"the amount is $839\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1988. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "37", "age_2": "38", "age_3": "35", "age_1": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "44", "answer_2": "24", "answer_1": "29", "answer_3": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 242:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $418 per year. It increases by 15% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "471.7", "answer_0": "458.7", "answer_1": "480.7", "answer_2": "355.3"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $169. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, October 01, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 448:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed December 24, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 01", "date_2": "December 30", "date_1": "December 30", "date_0": "January 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $29k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, November 27, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 3 grams of salt per serving and serves 6 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p0": "0.100", "p3": "0.050", "p1": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $962 per year. It increases by 25% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1159.5", "answer_1": "1202.5", "answer_0": "1178.5", "answer_2": "721.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = $371; priority = high.\n\nClaim to verify: \"the status is $371\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Sunday, July 19, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Thursday, April 02, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1980. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "46", "age_0": "44", "age_2": "45", "age_1": "43"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = 2026-09-16; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $842. A fee of $114 is charged, then a deposit of $26 arrives. A second fee of $57 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "669", "answer_3": "691", "answer_0": "697", "answer_1": "691"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 312 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "61", "answer_0": "80", "answer_1": "82", "answer_2": "61"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $586. A fee of $105 is charged, then a deposit of $37 arrives. A second fee of $52 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "441", "answer_3": "454", "answer_1": "476", "answer_0": "466"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, August 25, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1766 per year. It increases by 10% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1942.6", "answer_0": "1928.6", "answer_2": "1589.4", "answer_3": "1895.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $587. A fee of $71 is charged, then a deposit of $46 arrives. A second fee of $35 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "527", "answer_2": "550", "answer_1": "514", "answer_3": "500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=768, unit_price=857, discount_pct=0, region=EU, sku_total=1625 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "657408", "v2": "65817", "v0": "658176", "v1": "1625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 17:30 to 19:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h00", "dur_3": "1h00", "dur_0": "1h45", "dur_2": "2h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = $412; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $420. A fee of $124 is charged, then a deposit of $81 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "319", "answer_0": "315", "answer_1": "328", "answer_3": "330"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = $223; priority = low.\n\nClaim to verify: \"the status is $223\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed March 19, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 25", "date_1": "March 23", "date_2": "March 23", "date_3": "March 25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.286", "p0": "0.375", "p2": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Wednesday, February 11, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed January 24, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 04", "date_0": "February 05", "date_1": "February 02", "date_2": "February 07"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1960. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "62", "age_1": "63", "age_2": "65", "age_0": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=682, unit_price=151, discount_pct=0, region=EU, sku_total=833 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "103664", "v0": "102982", "v2": "10298", "v1": "833"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.125", "p0": "0.375", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Sunday, June 28, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Sunday, September 13, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 974:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p0": "0.625", "p3": "0.571", "p1": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 365:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.333", "p3": "0.636", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $254 per year. It increases by 25% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "288.5", "answer_0": "307.5", "answer_2": "190.5", "answer_1": "317.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $900. A fee of $53 is charged, then a deposit of $20 arrives. A second fee of $26 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "841", "answer_3": "859", "answer_2": "869", "answer_1": "812"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 461:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, February 28, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p0": "0.250", "p2": "0.125", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1996 per year. It increases by 10% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2183.6", "answer_3": "2149.6", "answer_2": "1796.4", "answer_1": "2195.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=984, unit_price=105, discount_pct=0, region=EU, sku_total=1089 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1089", "v2": "10332", "v3": "104304", "v0": "103320"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 119 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "188", "answer_3": "179", "answer_0": "192", "answer_1": "201"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, February 15, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $395. A fee of $126 is charged, then a deposit of $30 arrives. A second fee of $63 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "236", "answer_1": "221", "answer_2": "218", "answer_3": "214"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $855. A fee of $147 is charged, then a deposit of $83 arrives. A second fee of $73 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "718", "answer_2": "709", "answer_1": "745", "answer_3": "714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "51", "age_2": "53", "age_3": "50", "age_0": "52"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=440, unit_price=237, discount_pct=0, region=EU, sku_total=677 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "104280", "v3": "104720", "v1": "677", "v2": "10428"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = 2026-11-23; priority = high.\n\nClaim to verify: \"the status is 2026-11-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $858. A fee of $146 is charged, then a deposit of $51 arrives. A second fee of $73 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "666", "answer_2": "718", "answer_0": "690", "answer_3": "682"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $7k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1988. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "39", "age_2": "38", "age_1": "36", "age_0": "37"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_3": "24", "answer_0": "42", "answer_1": "22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = 2026-07-16; priority = high.\n\nClaim to verify: \"the date is 2026-07-16\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1346 per year. It increases by 20% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1076.8", "answer_3": "1575.2", "answer_1": "1615.2", "answer_0": "1568.2"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 152 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "39", "answer_1": "45", "answer_0": "42", "answer_2": "24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1990. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "32", "age_1": "33", "age_0": "34", "age_2": "35"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $203. A fee of $106 is charged, then a deposit of $25 arrives. A second fee of $53 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "69", "answer_1": "74", "answer_3": "42", "answer_2": "91"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 5 people. A shopper bought a 10 gram box and a 100 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = $355; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $715. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 08:15 to 19:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "10h45", "dur_2": "10h45", "dur_3": "9h45", "dur_0": "10h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p3": "0.400", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p0": "0.500", "p3": "0.429"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = 2026-08-06; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed November 21, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 03", "date_3": "December 02", "date_1": "November 30", "date_2": "December 05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1968. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "57", "age_3": "56", "age_2": "59", "age_0": "58"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "16", "answer_1": "16", "answer_0": "28", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 210:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "8", "answer_1": "12", "answer_0": "24", "answer_3": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed July 13, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 18", "date_2": "July 22", "date_0": "July 20", "date_3": "July 20"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=445, unit_price=522, discount_pct=0, region=EU, sku_total=967 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "232735", "v0": "232290", "v1": "967", "v2": "23229"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $730. A fee of $88 is charged, then a deposit of $70 arrives. A second fee of $44 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "672", "answer_0": "668", "answer_1": "683", "answer_3": "650"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 08:15 to 10:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h30", "dur_1": "2h00", "dur_3": "1h00", "dur_0": "1h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, June 03, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 135 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "50", "answer_0": "48", "answer_3": "57", "answer_1": "38"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = 2026-07-26; priority = high.\n\nClaim to verify: \"the amount is 2026-07-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Thursday, May 28, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, July 27, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "31", "answer_0": "66", "answer_3": "42", "answer_2": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 18:30 to 22:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h45", "dur_1": "4h00", "dur_3": "3h00", "dur_0": "3h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = 2026-07-02; priority = high.\n\nClaim to verify: \"the status is 2026-07-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.167", "p0": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $29k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, October 22, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "54", "answer_1": "29", "answer_2": "24", "answer_3": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.400", "p3": "0.500", "p0": "0.600", "p2": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $649. A fee of $63 is charged, then a deposit of $72 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "615", "answer_2": "617", "answer_1": "620", "answer_0": "627"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "40", "answer_3": "40", "answer_1": "45", "answer_0": "80"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.375", "p2": "0.125", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_2": "18", "answer_1": "22", "answer_0": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $518 per year. It increases by 20% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "621.6", "answer_2": "414.4", "answer_3": "624.6", "answer_0": "589.6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $715. A fee of $183 is charged, then a deposit of $81 arrives. A second fee of $91 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "522", "answer_1": "499", "answer_3": "551", "answer_2": "539"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "23", "answer_3": "56", "answer_2": "16", "answer_0": "72"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.625", "p2": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.600", "p2": "0.400", "p1": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=857, unit_price=487, discount_pct=0, region=EU, sku_total=1344 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "417359", "v1": "1344", "v3": "418216", "v2": "41735"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = 2026-03-05; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 80 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "175", "answer_2": "189", "answer_3": "188", "answer_1": "190"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=861, unit_price=331, discount_pct=0, region=EU, sku_total=1192 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "28499", "v1": "1192", "v3": "285852", "v0": "284991"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1843 per year. It increases by 15% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1566.55", "answer_1": "2119.45", "answer_3": "2125.45", "answer_0": "2096.45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 741:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=186, unit_price=231, discount_pct=0, region=EU, sku_total=417 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "417", "v2": "4296", "v3": "43152", "v0": "42966"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "30", "answer_0": "42", "answer_2": "28", "answer_3": "14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $840. A fee of $126 is charged, then a deposit of $21 arrives. A second fee of $63 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "653", "answer_0": "672", "answer_3": "657", "answer_1": "658"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 710:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.556", "p2": "0.400", "p0": "0.600", "p1": "0.400"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed November 22, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 02", "date_1": "November 30", "date_0": "December 02", "date_2": "December 04"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 285 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "78", "answer_1": "86", "answer_3": "66", "answer_0": "80"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 143 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "167", "answer_1": "171", "answer_2": "160", "answer_0": "162"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "19", "answer_0": "27", "answer_3": "12", "answer_2": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.111", "p0": "0.200", "p2": "0.800", "p1": "0.800"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 354 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "216", "answer_2": "223", "answer_3": "224", "answer_1": "197"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=387, unit_price=371, discount_pct=0, region=EU, sku_total=758 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "758", "v2": "14357", "v0": "143577", "v3": "143964"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 731:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 9 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Wednesday, June 03, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.500", "p3": "0.167", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_0": "33", "answer_1": "23", "answer_3": "15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, July 27, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = 2026-04-12; priority = high.\n\nClaim to verify: \"the status is 2026-04-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed October 20, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 02", "date_2": "November 04", "date_3": "October 31", "date_1": "October 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.250", "p2": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p0": "0.100", "p2": "0.050", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1926 per year. It increases by 15% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1637.1", "answer_0": "2186.9", "answer_3": "2146.9", "answer_1": "2214.9"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = 2026-03-12; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 9 grams of salt per serving and serves 7 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = $528; priority = low.\n\nClaim to verify: \"the date is $528\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, April 01, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Saturday, December 26, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 268:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.714", "p0": "0.286", "p2": "0.714"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 223 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "169", "answer_0": "189", "answer_2": "173", "answer_3": "179"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=219, unit_price=406, discount_pct=0, region=EU, sku_total=625 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "88914", "v1": "625", "v3": "88695", "v2": "8891"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "9", "answer_2": "15", "answer_1": "18", "answer_0": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = 2026-11-03; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=486, unit_price=457, discount_pct=0, region=EU, sku_total=943 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "943", "v3": "222588", "v0": "222102", "v2": "22210"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 106 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "150", "answer_1": "168", "answer_3": "148", "answer_2": "148"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed November 23, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 26", "date_1": "November 26", "date_3": "November 28", "date_2": "November 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_1": "35", "answer_3": "30", "answer_0": "60"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 365 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "278", "answer_1": "250", "answer_2": "255", "answer_0": "270"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = $69; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p0": "0.150", "p1": "0.200", "p3": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1962. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "64", "age_3": "61", "age_1": "62", "age_0": "63"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 317 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "105", "answer_2": "110", "answer_1": "120", "answer_3": "102"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1967. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "61", "age_0": "59", "age_2": "60", "age_1": "58"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_2": "32", "answer_3": "48", "answer_1": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $514 per year. It increases by 10% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "565.4", "answer_2": "462.6", "answer_0": "530.4", "answer_3": "498.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = $501; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 957:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $53. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = 2026-01-09; priority = low.\n\nClaim to verify: \"the status is 2026-01-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: owner = $224; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $21k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $5k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = $249; priority = high.\n\nClaim to verify: \"the date is $249\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $2000 per year. It increases by 20% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2369", "answer_1": "2400", "answer_0": "2352", "answer_2": "1600"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $13k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, April 11, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = $643; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = 2026-03-01; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, October 19, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 958:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=708, unit_price=955, discount_pct=0, region=EU, sku_total=1663 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1663", "v3": "676848", "v0": "676140", "v2": "67614"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 3 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 841:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.333", "p3": "0.167", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 940:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_2": "25", "answer_3": "15", "answer_1": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = $935; priority = low.\n\nClaim to verify: \"the status is $935\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 3 grams of salt per serving and serves 11 people. A shopper bought a 33 gram box and a 330 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 4 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = 2026-06-06; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed April 25, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 06", "date_0": "May 07", "date_2": "May 05", "date_1": "May 04"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1768 per year. It increases by 25% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2171", "answer_1": "2210", "answer_3": "2211", "answer_2": "1326"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 262:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 332 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "303", "answer_0": "288", "answer_2": "271", "answer_3": "302"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = $458; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "25", "answer_3": "10", "answer_1": "17", "answer_2": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.600", "p0": "0.636", "p2": "0.364", "p1": "0.364"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Tuesday, April 07, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1960. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "65", "age_3": "64", "age_2": "67", "age_0": "66"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Friday, July 03, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1964. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "59", "age_2": "61", "age_3": "58", "age_0": "60"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $32k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=482, unit_price=839, discount_pct=0, region=EU, sku_total=1321 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "40439", "v0": "404398", "v3": "404880", "v1": "1321"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = $168; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.500", "p3": "0.333", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, December 21, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1919 per year. It increases by 20% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1535.2", "answer_0": "2275.8", "answer_1": "2302.8", "answer_3": "2261.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 267 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_1": "64", "answer_2": "55", "answer_3": "84"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Wednesday, January 14, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = pending; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = $421; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.400", "p0": "0.600", "p1": "0.400", "p3": "0.556"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = $900; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 506:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Wednesday, July 01, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed September 16, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 24", "date_1": "September 22", "date_2": "September 26", "date_0": "September 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.615", "p3": "0.583", "p2": "0.385", "p1": "0.385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.667", "p1": "0.833", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=928, unit_price=586, discount_pct=0, region=EU, sku_total=1514 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "544736", "v0": "543808", "v1": "1514", "v2": "54380"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 56 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "70", "answer_3": "79", "answer_1": "88", "answer_2": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $533. A fee of $169 is charged, then a deposit of $30 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "310", "answer_2": "334", "answer_1": "297", "answer_3": "296"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.200", "p3": "0.050", "p2": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $5k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.250", "p3": "0.083", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.417", "p3": "0.364", "p2": "0.583", "p1": "0.583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.444", "p2": "0.444", "p0": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 410:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $890 per year. It increases by 20% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1068", "answer_0": "1035", "answer_2": "712", "answer_3": "999"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $703. A fee of $173 is charged, then a deposit of $81 arrives. A second fee of $86 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "498", "answer_1": "516", "answer_2": "547", "answer_0": "525"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = $939; priority = high.\n\nClaim to verify: \"the owner is $939\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $603. A fee of $58 is charged, then a deposit of $26 arrives. A second fee of $29 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "542", "answer_1": "554", "answer_3": "551", "answer_2": "523"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.385", "p1": "0.571", "p0": "0.429", "p2": "0.571"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 369 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "59", "answer_0": "50", "answer_1": "46", "answer_2": "45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = $88; priority = high.\n\nClaim to verify: \"the date is $88\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1981. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "42", "age_3": "41", "age_2": "44", "age_0": "43"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 374 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "123", "answer_3": "122", "answer_2": "113", "answer_0": "120"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 930:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = 2026-10-24; priority = high.\n\nClaim to verify: \"the owner is 2026-10-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = 2026-01-21; priority = low.\n\nClaim to verify: \"the amount is 2026-01-21\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $234. A fee of $161 is charged, then a deposit of $27 arrives. A second fee of $80 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_2": "-1", "answer_0": "20", "answer_1": "18"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1980. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "45", "age_2": "46", "age_1": "44", "age_3": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "32", "answer_3": "16", "answer_2": "16", "answer_1": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.444", "p1": "0.500", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.083", "p0": "0.167", "p1": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 229 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "101", "answer_0": "96", "answer_2": "111", "answer_1": "78"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Contract signed May 19, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 26", "date_3": "May 26", "date_1": "May 24", "date_2": "May 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed July 25, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 05", "date_0": "August 06", "date_1": "August 03", "date_2": "August 04"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1461 per year. It increases by 25% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1095.75", "answer_0": "1803.25", "answer_3": "1779.25", "answer_1": "1826.25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $667. A fee of $182 is charged, then a deposit of $84 arrives. A second fee of $91 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "491", "answer_0": "478", "answer_2": "456", "answer_3": "454"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $148. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.600", "p1": "0.333", "p2": "0.333", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = $776; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $5k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 318 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "12", "answer_1": "15", "answer_2": "16", "answer_3": "26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1045 per year. It increases by 25% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1259.25", "answer_2": "783.75", "answer_3": "1284.25", "answer_1": "1306.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.714", "p2": "0.250", "p0": "0.750"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = pending; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, April 01, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.750", "p0": "0.250", "p3": "0.143", "p2": "0.750"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed December 11, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 17", "date_2": "December 19", "date_0": "December 17", "date_1": "December 15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p0": "0.100", "p2": "0.050", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $798. A fee of $98 is charged, then a deposit of $83 arrives. A second fee of $49 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "734", "answer_2": "748", "answer_3": "742", "answer_1": "755"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 539:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.273", "p0": "0.333", "p2": "0.667", "p1": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p0": "0.500", "p3": "0.125", "p1": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.400", "p0": "0.455", "p1": "0.545", "p2": "0.545"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.467", "p0": "0.533", "p2": "0.467", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 247 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "312", "answer_3": "307", "answer_1": "340", "answer_0": "320"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $37k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 2 grams of salt per serving and serves 4 people. A shopper bought a 8 gram box and a 80 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $8k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "28", "answer_1": "31", "answer_3": "21", "answer_0": "49"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 285:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 367 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "78", "answer_3": "86", "answer_0": "75", "answer_2": "88"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=672, unit_price=246, discount_pct=0, region=EU, sku_total=918 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "16531", "v0": "165312", "v3": "164640", "v1": "918"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=937, unit_price=640, discount_pct=0, region=EU, sku_total=1577 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1577", "v3": "600617", "v2": "59968", "v0": "599680"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.286", "p1": "0.714", "p2": "0.714"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1962. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "65", "age_1": "62", "age_2": "64", "age_0": "63"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = 2026-06-24; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_0": "36", "answer_1": "21", "answer_3": "18"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 7 grams of salt per serving and serves 10 people. A shopper bought a 70 gram box and a 700 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.667", "p1": "0.300", "p0": "0.700", "p2": "0.300"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = $273; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: status = 2026-12-17; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 346:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=942, unit_price=329, discount_pct=0, region=EU, sku_total=1271 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1271", "v2": "30991", "v3": "308976", "v0": "309918"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.375", "p2": "0.125", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = 2026-10-21; priority = high.\n\nClaim to verify: \"the amount is 2026-10-21\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, January 28, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Saturday, January 10, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1966. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "57", "age_0": "58", "age_3": "60", "age_2": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=284, unit_price=883, discount_pct=0, region=EU, sku_total=1167 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1167", "v0": "250772", "v3": "251056", "v2": "25077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 224:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed December 20, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 28", "date_3": "December 30", "date_0": "December 30", "date_1": "December 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $443. A fee of $78 is charged, then a deposit of $48 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "360", "answer_2": "381", "answer_3": "394", "answer_0": "374"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.667", "p0": "0.714", "p2": "0.286", "p1": "0.286"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=668, unit_price=131, discount_pct=0, region=EU, sku_total=799 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "87508", "v1": "799", "v2": "8750", "v3": "86840"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = $557; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1979. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "45", "age_3": "47", "age_1": "44", "age_2": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=808, unit_price=411, discount_pct=0, region=EU, sku_total=1219 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "332088", "v3": "332896", "v2": "33208", "v1": "1219"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "25", "answer_3": "28", "answer_2": "21", "answer_0": "49"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1327 per year. It increases by 25% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1631.75", "answer_1": "1658.75", "answer_2": "995.25", "answer_3": "1622.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = 2026-02-09; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.556", "p0": "0.444", "p3": "0.375", "p2": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $215. A fee of $187 is charged, then a deposit of $81 arrives. A second fee of $93 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "23", "answer_3": "-8", "answer_0": "16", "answer_1": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 708:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed October 05, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 14", "date_1": "October 14", "date_0": "October 16", "date_3": "October 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $11k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, February 23, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.600", "p2": "0.364", "p0": "0.636", "p1": "0.364"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = 2026-10-17; priority = low.\n\nClaim to verify: \"the amount is 2026-10-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed June 26, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 08", "date_0": "July 06", "date_1": "July 02", "date_3": "July 04"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = pending; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.600", "p1": "0.364", "p0": "0.636", "p2": "0.364"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.250", "p1": "0.417", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 283 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "27", "answer_0": "30", "answer_2": "18", "answer_1": "35"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "35", "answer_3": "49", "answer_2": "28", "answer_0": "77"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p1": "0.250", "p0": "0.200", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=865, unit_price=949, discount_pct=0, region=EU, sku_total=1814 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "820020", "v2": "82088", "v1": "1814", "v0": "820885"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = 2026-07-03; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 13:30 to 20:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "7h15", "dur_2": "6h45", "dur_3": "6h15", "dur_0": "6h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $705. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Saturday, September 05, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p2": "0.100", "p1": "0.200", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.462", "p2": "0.462", "p3": "0.500", "p0": "0.538"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 9 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 350 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "64", "answer_1": "84", "answer_3": "47", "answer_2": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=720, unit_price=338, discount_pct=0, region=EU, sku_total=1058 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "242640", "v1": "1058", "v0": "243360", "v2": "24336"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_3": "16", "answer_1": "26", "answer_2": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = $369; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1963. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "63", "age_1": "60", "age_2": "62", "age_0": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p1": "0.200", "p3": "0.050", "p2": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1973. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "54", "age_3": "51", "age_1": "52", "age_0": "53"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = $474; priority = low.\n\nClaim to verify: \"the status is $474\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $626. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = 2026-12-03; priority = high.\n\nClaim to verify: \"the status is 2026-12-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 386 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "153", "answer_2": "149", "answer_1": "160", "answer_0": "140"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "30", "answer_2": "18", "answer_3": "12", "answer_1": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $397. A fee of $32 is charged, then a deposit of $50 arrives. A second fee of $16 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "403", "answer_3": "390", "answer_0": "399", "answer_1": "405"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $31k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 227:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $360. A fee of $65 is charged, then a deposit of $25 arrives. A second fee of $32 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "288", "answer_1": "261", "answer_2": "293", "answer_3": "313"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = 2026-02-04; priority = low.\n\nClaim to verify: \"the amount is 2026-02-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1078 per year. It increases by 20% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "862.4", "answer_1": "1293.6", "answer_3": "1287.6", "answer_0": "1267.6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $759. A fee of $116 is charged, then a deposit of $80 arrives. A second fee of $58 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "641", "answer_0": "665", "answer_2": "658", "answer_1": "641"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed May 11, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 18", "date_2": "May 22", "date_3": "May 20", "date_0": "May 20"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.150", "p0": "0.100", "p2": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $340. A fee of $124 is charged, then a deposit of $70 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "210", "answer_1": "230", "answer_0": "224", "answer_3": "230"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = $649; priority = low.\n\nClaim to verify: \"the amount is $649\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p0": "0.667", "p3": "0.167", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.500", "p2": "0.333", "p1": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = $903; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 7 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.500", "p0": "0.667", "p1": "0.833"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=586, unit_price=131, discount_pct=0, region=EU, sku_total=717 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "7676", "v0": "76766", "v3": "76180", "v1": "717"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.667", "p3": "0.200", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1594 per year. It increases by 15% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1833.1", "answer_2": "1354.9", "answer_3": "1783.1", "answer_0": "1790.1"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = pending; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=597, unit_price=658, discount_pct=0, region=EU, sku_total=1255 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "393423", "v0": "392826", "v2": "39282", "v1": "1255"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 256 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_3": "96", "answer_2": "86", "answer_1": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "25", "answer_2": "20", "answer_0": "45", "answer_1": "25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=950, unit_price=449, discount_pct=0, region=EU, sku_total=1399 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "426550", "v3": "427500", "v2": "42655", "v1": "1399"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.273", "p2": "0.727", "p3": "0.200", "p1": "0.727"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "30", "answer_2": "15", "answer_0": "45", "answer_1": "21"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=501, unit_price=638, discount_pct=0, region=EU, sku_total=1139 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "320139", "v1": "1139", "v2": "31963", "v0": "319638"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p0": "0.333", "p1": "0.667", "p3": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.667", "p0": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 9 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, December 14, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $19k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 123 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "96", "answer_2": "77", "answer_0": "84", "answer_3": "102"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "20", "answer_0": "40", "answer_3": "20", "answer_1": "25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 186 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "252", "answer_1": "232", "answer_3": "256", "answer_2": "243"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $36k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $405. A fee of $171 is charged, then a deposit of $76 arrives. A second fee of $85 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "203", "answer_3": "203", "answer_0": "225", "answer_2": "255"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 07:00 to 19:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "11h15", "dur_2": "12h00", "dur_1": "12h15", "dur_0": "11h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $719. A fee of $175 is charged, then a deposit of $57 arrives. A second fee of $87 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "491", "answer_3": "518", "answer_2": "518", "answer_0": "514"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p3": "0.050", "p1": "0.150", "p0": "0.100"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $21k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.364", "p3": "0.600", "p2": "0.364", "p0": "0.636"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 7 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $894. A fee of $101 is charged, then a deposit of $68 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "818", "answer_3": "826", "answer_0": "811", "answer_1": "783"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.125", "p1": "0.375", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $301. A fee of $166 is charged, then a deposit of $58 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "127", "answer_0": "110", "answer_1": "92", "answer_3": "123"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 540:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, November 24, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Monday, October 19, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $463 per year. It increases by 25% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "550.75", "answer_3": "529.75", "answer_2": "347.25", "answer_1": "578.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $775 per year. It increases by 25% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "933.75", "answer_1": "968.75", "answer_2": "581.25", "answer_3": "960.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.150", "p3": "0.050", "p0": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.538", "p1": "0.462", "p2": "0.462", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_1": "16", "answer_0": "28", "answer_2": "14"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 7 grams of salt per serving and serves 6 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, December 15, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1964. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "61", "age_2": "63", "age_0": "62", "age_3": "64"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_2": "35", "answer_1": "40", "answer_0": "70"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 941:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 405:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 5 people. A shopper bought a 10 gram box and a 100 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $867. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = 2026-08-05; priority = high.\n\nClaim to verify: \"the owner is 2026-08-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=552, unit_price=390, discount_pct=0, region=EU, sku_total=942 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "942", "v3": "214728", "v2": "21528", "v0": "215280"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 248 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "123", "answer_0": "105", "answer_3": "99", "answer_2": "103"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=702, unit_price=934, discount_pct=0, region=EU, sku_total=1636 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "654966", "v1": "1636", "v0": "655668", "v2": "65566"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $682 per year. It increases by 25% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "852.5", "answer_3": "803.5", "answer_2": "511.5", "answer_0": "818.5"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $425. A fee of $72 is charged, then a deposit of $62 arrives. A second fee of $36 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "376", "answer_0": "379", "answer_3": "358", "answer_1": "402"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 67 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "150", "answer_1": "149", "answer_2": "126", "answer_0": "144"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.167", "p1": "0.833", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Wednesday, November 18, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $514. A fee of $86 is charged, then a deposit of $82 arrives. A second fee of $43 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "447", "answer_1": "447", "answer_0": "467", "answer_3": "482"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 4 grams of salt per serving and serves 10 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1990. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "34", "age_3": "33", "age_2": "36", "age_0": "35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = open; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1976. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "49", "age_3": "52", "age_0": "50", "age_2": "51"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 14:00 to 21:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "7h30", "dur_3": "6h45", "dur_0": "7h00", "dur_1": "7h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p0": "0.150", "p3": "0.050", "p2": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.417", "p2": "0.250", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: date = ops; priority = low.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $318. A fee of $51 is charged, then a deposit of $66 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "333", "answer_3": "321", "answer_0": "308", "answer_1": "286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $802 per year. It increases by 5% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "761.9", "answer_1": "842.1", "answer_0": "827.1", "answer_3": "794.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $29k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_1": "24", "answer_3": "18", "answer_0": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p3": "0.200", "p1": "0.667", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.400", "p0": "0.500", "p2": "0.500", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 962:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 10:15 to 18:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "6h45", "dur_0": "7h30", "dur_1": "7h45", "dur_2": "7h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.167", "p3": "0.167", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = $32; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.200", "p2": "0.667", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $264. A fee of $111 is charged, then a deposit of $73 arrives. A second fee of $55 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "158", "answer_3": "147", "answer_1": "166", "answer_0": "171"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.250", "p0": "0.375", "p3": "0.125"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Thursday, October 22, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.500", "p2": "0.167", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $542 per year. It increases by 5% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "514.9", "answer_0": "545.1", "answer_1": "569.1", "answer_3": "571.1"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = 2026-04-27; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "62", "age_0": "64", "age_1": "63", "age_2": "65"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "20", "answer_1": "11", "answer_2": "8", "answer_3": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 3 grams of salt per serving and serves 6 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.636", "p1": "0.333", "p2": "0.333", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "8", "answer_0": "28", "answer_1": "22", "answer_2": "20"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "51", "answer_3": "24", "answer_0": "72", "answer_2": "48"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed September 16, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 26", "date_0": "September 28", "date_2": "September 26", "date_1": "September 24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 961:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.167", "p2": "0.333", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.167", "p3": "0.083", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $698 per year. It increases by 10% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "724.8", "answer_2": "628.2", "answer_1": "767.8", "answer_3": "693.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 5 grams of salt per serving and serves 4 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = 2026-04-12; priority = high.\n\nClaim to verify: \"the status is 2026-04-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, October 24, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=820, unit_price=606, discount_pct=0, region=EU, sku_total=1426 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "49692", "v1": "1426", "v0": "496920", "v3": "496100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p0": "0.400", "p2": "0.600", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = $712; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, July 25, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = 2026-09-18; priority = high.\n\nClaim to verify: \"the owner is 2026-09-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1989. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "33", "age_0": "35", "age_1": "34", "age_2": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $617. A fee of $87 is charged, then a deposit of $55 arrives. A second fee of $43 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "564", "answer_0": "542", "answer_3": "554", "answer_2": "562"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 903:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 609:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, May 06, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=388, unit_price=787, discount_pct=0, region=EU, sku_total=1175 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "30535", "v1": "1175", "v3": "304968", "v0": "305356"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 8 grams of salt per serving and serves 8 people. A shopper bought a 64 gram box and a 640 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 103 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "112", "answer_3": "97", "answer_1": "103", "answer_0": "100"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 7 grams of salt per serving and serves 4 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1494 per year. It increases by 10% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1611.4", "answer_2": "1344.6", "answer_3": "1595.4", "answer_1": "1643.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $214. A fee of $152 is charged, then a deposit of $65 arrives. A second fee of $76 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "51", "answer_3": "39", "answer_2": "76", "answer_1": "73"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "21", "answer_0": "27", "answer_2": "18", "answer_3": "9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 73 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "90", "answer_1": "109", "answer_3": "97", "answer_2": "106"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $472. A fee of $52 is charged, then a deposit of $48 arrives. A second fee of $26 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "464", "answer_2": "421", "answer_1": "418", "answer_0": "442"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=978, unit_price=869, discount_pct=0, region=EU, sku_total=1847 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "84988", "v0": "849882", "v1": "1847", "v3": "848904"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $354. A fee of $57 is charged, then a deposit of $81 arrives. A second fee of $28 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "327", "answer_3": "332", "answer_2": "354", "answer_0": "350"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.125", "p1": "0.500", "p0": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $658. A fee of $163 is charged, then a deposit of $46 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "460", "answer_3": "490", "answer_2": "486", "answer_1": "433"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.462", "p2": "0.462", "p0": "0.538", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $518. A fee of $152 is charged, then a deposit of $29 arrives. A second fee of $76 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "319", "answer_3": "301", "answer_2": "346", "answer_1": "316"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Thursday, July 16, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 6 grams of salt per serving and serves 5 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p0": "0.667", "p2": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 107 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "73", "answer_3": "72", "answer_1": "97", "answer_0": "84"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 809:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "45", "answer_3": "20", "answer_2": "25", "answer_1": "29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.375", "p1": "0.500", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed August 06, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 14", "date_0": "August 14", "date_2": "August 12", "date_1": "August 12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.333", "p0": "0.500", "p1": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $653. A fee of $74 is charged, then a deposit of $63 arrives. A second fee of $37 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "605", "answer_2": "580", "answer_1": "628", "answer_3": "588"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = 2026-07-21; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 5 grams of salt per serving and serves 3 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 349:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = 2026-04-07; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = $944; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $523. A fee of $163 is charged, then a deposit of $64 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "343", "answer_3": "367", "answer_1": "332", "answer_2": "371"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.615", "p2": "0.385", "p3": "0.583", "p1": "0.385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 795:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 06:30 to 14:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "8h00", "dur_3": "7h00", "dur_0": "7h30", "dur_1": "8h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1976. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "49", "age_0": "50", "age_3": "48", "age_2": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $955 per year. It increases by 10% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1056.5", "answer_2": "859.5", "answer_1": "1050.5", "answer_0": "1029.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "8", "answer_2": "8", "answer_1": "10", "answer_0": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = open; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p1": "0.500", "p3": "0.467"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 273 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "117", "answer_1": "91", "answer_0": "108", "answer_3": "119"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.167", "p0": "0.333", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.125", "p1": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = 2026-02-02; priority = low.\n\nClaim to verify: \"the date is 2026-02-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $38k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed June 18, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 25", "date_2": "June 23", "date_0": "June 25", "date_1": "June 23"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.286", "p1": "0.625", "p2": "0.625", "p0": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 255 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "83", "answer_2": "97", "answer_0": "80", "answer_1": "65"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 07:00 to 19:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "11h45", "dur_1": "12h45", "dur_0": "12h30", "dur_2": "13h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 07:00 to 16:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "8h45", "dur_1": "9h00", "dur_2": "9h15", "dur_3": "8h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, October 26, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.333", "p1": "0.417", "p3": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $819. A fee of $58 is charged, then a deposit of $56 arrives. A second fee of $29 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "786", "answer_2": "796", "answer_3": "777", "answer_0": "788"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p1": "0.200", "p0": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $799. A fee of $197 is charged, then a deposit of $67 arrives. A second fee of $98 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "553", "answer_1": "543", "answer_0": "571", "answer_2": "559"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = 2026-07-27; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p3": "0.125", "p1": "0.375", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = 2026-01-04; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = $555; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $578. A fee of $88 is charged, then a deposit of $22 arrives. A second fee of $44 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "474", "answer_0": "468", "answer_2": "483", "answer_3": "452"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.571", "p1": "0.375", "p2": "0.375", "p0": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = closed; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1979. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "45", "age_2": "47", "age_3": "44", "age_0": "46"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.400", "p3": "0.500", "p1": "0.400", "p0": "0.600"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1976. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "48", "age_2": "51", "age_0": "50", "age_1": "49"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=969, unit_price=590, discount_pct=0, region=EU, sku_total=1559 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "570741", "v0": "571710", "v2": "57171", "v1": "1559"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = $393; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed October 17, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 22", "date_0": "October 21", "date_1": "October 20", "date_2": "October 19"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "64", "age_3": "66", "age_2": "65", "age_1": "63"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.538", "p0": "0.571", "p1": "0.429", "p2": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $702 per year. It increases by 15% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "794.3", "answer_2": "596.7", "answer_1": "807.3", "answer_3": "780.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = 2026-03-17; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 288 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "261", "answer_1": "287", "answer_0": "280", "answer_3": "266"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 4 grams of salt per serving and serves 3 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.200", "p0": "0.150", "p2": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $689. A fee of $106 is charged, then a deposit of $39 arrives. A second fee of $53 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "549", "answer_3": "553", "answer_2": "593", "answer_0": "569"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $321 per year. It increases by 10% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "329.1", "answer_1": "353.1", "answer_2": "288.9", "answer_3": "305.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed January 16, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 27", "date_2": "January 27", "date_0": "January 29", "date_1": "January 25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 252 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "58", "answer_0": "54", "answer_2": "50", "answer_3": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1695 per year. It increases by 15% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1949.25", "answer_3": "1905.25", "answer_2": "1440.75", "answer_0": "1924.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 414:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, March 02, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "20", "answer_2": "18", "answer_0": "24", "answer_3": "6"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = 2026-04-24; priority = low.\n\nClaim to verify: \"the amount is 2026-04-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "15", "answer_2": "15", "answer_0": "30", "answer_1": "20"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1990. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "36", "age_2": "37", "age_1": "35", "age_3": "34"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 853:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = 2026-06-01; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = $749; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.300", "p0": "0.700", "p2": "0.300", "p3": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $669. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1835 per year. It increases by 25% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1376.25", "answer_0": "2256.75", "answer_3": "2218.75", "answer_1": "2293.75"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 8 grams of salt per serving and serves 5 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $295. A fee of $78 is charged, then a deposit of $84 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "257", "answer_2": "278", "answer_3": "276", "answer_0": "262"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $378. A fee of $133 is charged, then a deposit of $90 arrives. A second fee of $66 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "241", "answer_2": "298", "answer_1": "241", "answer_0": "269"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.286", "p3": "0.167", "p1": "0.714", "p2": "0.714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1688 per year. It increases by 15% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1908.2", "answer_2": "1434.8", "answer_1": "1941.2", "answer_0": "1923.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1860 per year. It increases by 5% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1767", "answer_0": "1929", "answer_1": "1953", "answer_3": "1895"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1959. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "66", "age_0": "67", "age_3": "65", "age_2": "68"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1983. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "41", "age_3": "40", "age_0": "42", "age_2": "43"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = 2026-01-09; priority = low.\n\nClaim to verify: \"the owner is 2026-01-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 217 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "22", "answer_3": "18", "answer_2": "36", "answer_0": "24"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = $72; priority = low.\n\nClaim to verify: \"the amount is $72\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $788. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1985. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "39", "age_3": "42", "age_0": "40", "age_2": "41"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $682. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=264, unit_price=361, discount_pct=0, region=EU, sku_total=625 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "95568", "v1": "625", "v0": "95304", "v2": "9530"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "14", "answer_0": "35", "answer_3": "21", "answer_1": "17"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=437, unit_price=494, discount_pct=0, region=EU, sku_total=931 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "21587", "v0": "215878", "v1": "931", "v3": "215441"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=769, unit_price=696, discount_pct=0, region=EU, sku_total=1465 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "53522", "v3": "534455", "v1": "1465", "v0": "535224"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 352 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "96", "answer_2": "86", "answer_3": "115", "answer_1": "104"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $16k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $656 per year. It increases by 20% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "773.2", "answer_3": "798.2", "answer_2": "524.8", "answer_1": "787.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $809. A fee of $102 is charged, then a deposit of $26 arrives. A second fee of $51 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "663", "answer_1": "664", "answer_0": "682", "answer_3": "705"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = closed; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $40k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 5 grams of salt per serving and serves 12 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, May 08, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_0": "18", "answer_1": "10", "answer_2": "6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $123. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = $862; priority = low.\n\nClaim to verify: \"the owner is $862\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1981. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "43", "age_0": "44", "age_2": "45", "age_3": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 13:15 to 14:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "0h30", "dur_2": "1h30", "dur_1": "1h30", "dur_0": "1h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, May 04, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.250", "p0": "0.167", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1889 per year. It increases by 10% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2041.9", "answer_1": "2077.9", "answer_0": "2066.9", "answer_2": "1700.1"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Contract signed February 26, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 10", "date_3": "March 08", "date_2": "March 08", "date_1": "March 06"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 63 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "94", "answer_2": "80", "answer_1": "91", "answer_0": "96"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = $732; priority = high.\n\nClaim to verify: \"the amount is $732\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=216, unit_price=513, discount_pct=0, region=EU, sku_total=729 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "110808", "v1": "729", "v3": "110592", "v2": "11080"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 263 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "-8", "answer_1": "-6", "answer_3": "5", "answer_0": "12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p3": "0.125", "p0": "0.250", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.125", "p1": "0.375", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = $490; priority = low.\n\nClaim to verify: \"the status is $490\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "12", "answer_0": "24", "answer_3": "16", "answer_2": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.273", "p1": "0.667", "p0": "0.333", "p2": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 3 grams of salt per serving and serves 9 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 140 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "84", "answer_0": "80", "answer_1": "73", "answer_2": "87"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 11 people. A shopper bought a 66 gram box and a 660 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = 2026-11-12; priority = high.\n\nClaim to verify: \"the amount is 2026-11-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1870 per year. It increases by 15% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2090.5", "answer_0": "2117.5", "answer_2": "1589.5", "answer_1": "2150.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 71 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "43", "answer_0": "40", "answer_3": "25", "answer_2": "34"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "20", "answer_0": "28", "answer_2": "8", "answer_1": "13"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = $482; priority = low.\n\nClaim to verify: \"the status is $482\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 234 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "79", "answer_0": "63", "answer_2": "77", "answer_1": "51"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $257. A fee of $179 is charged, then a deposit of $52 arrives. A second fee of $89 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_1": "27", "answer_0": "41", "answer_3": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p1": "0.600", "p3": "0.333", "p2": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed October 06, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 13", "date_1": "October 11", "date_0": "October 13", "date_2": "October 15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 11 people. A shopper bought a 33 gram box and a 330 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.545", "p1": "0.417", "p0": "0.583", "p2": "0.417"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An account starts at $622. A fee of $75 is charged, then a deposit of $23 arrives. A second fee of $37 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "563", "answer_0": "533", "answer_3": "554", "answer_1": "537"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 326 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "369", "answer_0": "378", "answer_3": "386", "answer_1": "364"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_3": "36", "answer_0": "60", "answer_1": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, December 02, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p1": "0.615", "p0": "0.385", "p2": "0.615"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1277 per year. It increases by 15% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1085.45", "answer_3": "1435.55", "answer_0": "1443.55", "answer_1": "1468.55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1983. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "41", "age_0": "43", "age_2": "44", "age_1": "42"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 342 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "57", "answer_2": "57", "answer_0": "72", "answer_1": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p0": "0.500", "p3": "0.455"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 602:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1969. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "56", "age_0": "57", "age_3": "59", "age_2": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 7 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 361 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "172", "answer_2": "170", "answer_3": "144", "answer_0": "160"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $18k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 6 grams of salt per serving and serves 10 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 383:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $291. A fee of $66 is charged, then a deposit of $31 arrives. A second fee of $33 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "198", "answer_2": "229", "answer_0": "223", "answer_3": "245"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 380 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_3": "90", "answer_1": "93", "answer_2": "75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $811. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = $206; priority = high.\n\nClaim to verify: \"the amount is $206\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: status = 2026-08-26; priority = low.\n\nClaim to verify: \"the status is 2026-08-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: owner = 2026-10-12; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=596, unit_price=113, discount_pct=0, region=EU, sku_total=709 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "709", "v3": "67944", "v2": "6734", "v0": "67348"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=986, unit_price=554, discount_pct=0, region=EU, sku_total=1540 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "546244", "v1": "1540", "v3": "545258", "v2": "54624"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p2": "0.571", "p3": "0.333", "p0": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.333", "p1": "0.500", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $843 per year. It increases by 10% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "913.3", "answer_2": "758.7", "answer_1": "927.3", "answer_0": "881.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "70", "answer_1": "40", "answer_3": "35", "answer_2": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.167", "p3": "0.083", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $258. A fee of $50 is charged, then a deposit of $64 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "236", "answer_0": "247", "answer_1": "271", "answer_3": "255"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=251, unit_price=438, discount_pct=0, region=EU, sku_total=689 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "109938", "v2": "10993", "v1": "689", "v3": "110189"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $252. A fee of $48 is charged, then a deposit of $66 arrives. A second fee of $24 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "252", "answer_3": "260", "answer_0": "246", "answer_1": "241"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p3": "0.083", "p1": "0.250", "p0": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = $700; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1600 per year. It increases by 5% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1680", "answer_3": "1652", "answer_0": "1639", "answer_2": "1520"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Contract signed April 20, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 30", "date_1": "April 28", "date_0": "April 30", "date_2": "April 28"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = 2026-12-12; priority = low.\n\nClaim to verify: \"the amount is 2026-12-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $17k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 426:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, November 02, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.083", "p2": "0.167", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $825. A fee of $140 is charged, then a deposit of $20 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "627", "answer_0": "635", "answer_1": "628", "answer_3": "659"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $362. A fee of $70 is charged, then a deposit of $28 arrives. A second fee of $35 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "285", "answer_3": "295", "answer_1": "295", "answer_2": "269"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = 2026-03-19; priority = low.\n\nClaim to verify: \"the amount is 2026-03-19\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.455", "p2": "0.500", "p1": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=465, unit_price=643, discount_pct=0, region=EU, sku_total=1108 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "29899", "v3": "298530", "v0": "298995", "v1": "1108"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $451. A fee of $34 is charged, then a deposit of $82 arrives. A second fee of $17 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "482", "answer_1": "459", "answer_3": "497", "answer_2": "487"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $366. A fee of $63 is charged, then a deposit of $75 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "357", "answer_2": "319", "answer_0": "347", "answer_1": "357"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 513:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 240 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "47", "answer_2": "34", "answer_0": "40", "answer_3": "47"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 453:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = $524; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, February 28, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $873. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1970. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "54", "age_1": "53", "age_2": "55", "age_3": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An account starts at $360. A fee of $155 is charged, then a deposit of $54 arrives. A second fee of $77 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "160", "answer_1": "207", "answer_2": "164", "answer_0": "182"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Wednesday, July 22, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 370:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 710:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = 2026-08-08; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "19", "answer_3": "35", "answer_0": "49", "answer_2": "14"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $498. A fee of $85 is charged, then a deposit of $20 arrives. A second fee of $42 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "388", "answer_0": "391", "answer_2": "394", "answer_1": "382"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.429", "p0": "0.571", "p3": "0.500", "p2": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.333", "p3": "0.083", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A shift runs from 09:45 to 22:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "12h00", "dur_0": "12h15", "dur_1": "13h00", "dur_2": "12h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = 2026-03-03; priority = high.\n\nClaim to verify: \"the amount is 2026-03-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "21", "answer_3": "15", "answer_1": "11", "answer_2": "6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $19k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "10", "answer_3": "8", "answer_0": "16", "answer_2": "8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 6 grams of salt per serving and serves 10 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $273. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p2": "0.250", "p3": "0.083", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed February 15, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 22", "date_0": "February 20", "date_3": "February 22", "date_1": "February 20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed August 22, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 02", "date_3": "September 01", "date_2": "September 04", "date_1": "August 30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $898. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "49", "answer_3": "14", "answer_1": "37", "answer_2": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.222", "p2": "0.778", "p1": "0.778"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p3": "0.455", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.417", "p1": "0.417", "p0": "0.583", "p3": "0.545"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 319 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "50", "answer_3": "57", "answer_1": "74", "answer_0": "64"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 278 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "202", "answer_0": "196", "answer_1": "185", "answer_2": "177"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $771 per year. It increases by 20% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "616.8", "answer_0": "878.2", "answer_3": "851.2", "answer_1": "925.2"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $724. A fee of $187 is charged, then a deposit of $52 arrives. A second fee of $93 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "489", "answer_1": "505", "answer_3": "466", "answer_0": "496"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = 2026-02-17; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1980. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "44", "age_2": "47", "age_1": "45", "age_0": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $10k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 119 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "214", "answer_0": "210", "answer_2": "217", "answer_1": "190"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1980. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "47", "age_3": "44", "age_0": "46", "age_1": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.500", "p3": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=117, unit_price=403, discount_pct=0, region=EU, sku_total=520 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "4715", "v3": "47268", "v0": "47151", "v1": "520"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = 2026-10-07; priority = low.\n\nClaim to verify: \"the amount is 2026-10-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $38k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 206 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "96", "answer_1": "108", "answer_2": "84", "answer_3": "78"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = 2026-01-23; priority = low.\n\nClaim to verify: \"the date is 2026-01-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $848. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "49", "answer_2": "21", "answer_3": "28", "answer_1": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.500", "p0": "0.333", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = $711; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.050", "p1": "0.150", "p0": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = $360; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $284. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "42", "answer_2": "36", "answer_3": "36", "answer_0": "72"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 2 grams of salt per serving and serves 6 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Sunday, May 10, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.667", "p1": "0.833", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 8 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p2": "0.500", "p3": "0.167", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.700", "p3": "0.222", "p0": "0.300", "p1": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1963. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "63", "age_0": "62", "age_3": "60", "age_1": "61"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $558 per year. It increases by 25% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "638.5", "answer_1": "697.5", "answer_2": "418.5", "answer_0": "668.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $431 per year. It increases by 25% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "538.75", "answer_3": "462.75", "answer_0": "499.75", "answer_2": "323.25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "23", "answer_3": "28", "answer_2": "16", "answer_0": "44"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, April 27, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $614. A fee of $189 is charged, then a deposit of $73 arrives. A second fee of $94 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "432", "answer_2": "434", "answer_1": "384", "answer_0": "404"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1959. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "65", "age_3": "68", "age_2": "67", "age_0": "66"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.417", "p0": "0.583", "p1": "0.417", "p3": "0.545"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_1": "33", "answer_0": "63", "answer_2": "28"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $421. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 332 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "102", "answer_3": "124", "answer_1": "123", "answer_0": "108"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1979. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "45", "age_1": "44", "age_3": "43", "age_2": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $233. A fee of $117 is charged, then a deposit of $35 arrives. A second fee of $58 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "80", "answer_0": "93", "answer_1": "90", "answer_3": "82"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1978. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "46", "age_2": "47", "age_1": "45", "age_3": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 823:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, October 27, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = closed; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed January 10, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 20", "date_2": "January 25", "date_0": "January 23", "date_3": "January 22"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = 2026-10-22; priority = low.\n\nClaim to verify: \"the amount is 2026-10-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p0": "0.100", "p1": "0.150", "p3": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 9 grams of salt per serving and serves 12 people. A shopper bought a 108 gram box and a 1080 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "70", "answer_3": "42", "answer_1": "34", "answer_2": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Thursday, October 22, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 437:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = 2026-08-26; priority = high.\n\nClaim to verify: \"the date is 2026-08-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=988, unit_price=212, discount_pct=0, region=EU, sku_total=1200 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "209456", "v2": "20945", "v3": "210444", "v1": "1200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 7 grams of salt per serving and serves 6 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $712 per year. It increases by 5% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "747.6", "answer_2": "676.4", "answer_3": "694.6", "answer_0": "706.6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.375", "p3": "0.125", "p1": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $563. A fee of $169 is charged, then a deposit of $83 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "378", "answer_3": "414", "answer_0": "393", "answer_1": "374"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, January 20, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1981. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "43", "age_0": "45", "age_2": "46", "age_1": "44"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 229 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "279", "answer_3": "302", "answer_1": "282", "answer_0": "288"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 82 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "30", "answer_2": "46", "answer_0": "36", "answer_1": "55"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, August 02, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = 2026-12-10; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=435, unit_price=331, discount_pct=0, region=EU, sku_total=766 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "143550", "v1": "766", "v0": "143985", "v2": "14398"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Record fields: status = pending; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=532, unit_price=756, discount_pct=0, region=EU, sku_total=1288 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "402192", "v3": "401660", "v2": "40219", "v1": "1288"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p0": "0.250", "p3": "0.125", "p1": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Thursday, July 09, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=195, unit_price=408, discount_pct=0, region=EU, sku_total=603 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "79560", "v2": "7956", "v1": "603", "v3": "79365"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_1": "23", "answer_3": "30", "answer_0": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1958. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "66", "age_2": "67", "age_1": "65", "age_3": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 256:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 326 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "31", "answer_1": "59", "answer_0": "40", "answer_2": "27"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $350 per year. It increases by 20% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "420", "answer_2": "280", "answer_0": "378", "answer_3": "358"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 6 grams of salt per serving and serves 9 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 247:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, July 17, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.333", "p3": "0.167", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, May 14, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $5k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 420:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $234. A fee of $180 is charged, then a deposit of $47 arrives. A second fee of $90 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "11", "answer_1": "31", "answer_2": "31", "answer_3": "21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "25", "answer_2": "10", "answer_3": "15", "answer_1": "13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1215 per year. It increases by 15% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1032.75", "answer_1": "1397.25", "answer_0": "1352.25", "answer_3": "1377.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1456 per year. It increases by 10% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1589.6", "answer_1": "1601.6", "answer_2": "1310.4", "answer_0": "1573.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $512. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.222", "p0": "0.300", "p2": "0.700", "p1": "0.700"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "39", "answer_2": "32", "answer_3": "56", "answer_0": "88"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.286", "p1": "0.625", "p2": "0.625", "p0": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = closed; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: status = open; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed March 23, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 02", "date_3": "March 31", "date_0": "March 31", "date_1": "March 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $349. A fee of $134 is charged, then a deposit of $87 arrives. A second fee of $67 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "208", "answer_2": "241", "answer_0": "235", "answer_3": "244"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed December 22, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 27", "date_1": "December 25", "date_0": "December 25", "date_2": "December 23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = $157; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $290. A fee of $30 is charged, then a deposit of $59 arrives. A second fee of $15 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "304", "answer_2": "328", "answer_3": "321", "answer_1": "298"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.333", "p0": "0.250", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "40", "answer_1": "44", "answer_0": "72", "answer_3": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Saturday, December 26, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 07:15 to 08:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "0h45", "dur_2": "1h00", "dur_3": "0h15", "dur_1": "1h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $593. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 859:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p2": "0.625", "p3": "0.286", "p0": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_1": "19", "answer_0": "27", "answer_2": "15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = 2026-05-01; priority = high.\n\nClaim to verify: \"the owner is 2026-05-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.200", "p1": "0.250", "p2": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.667", "p1": "0.300", "p2": "0.300", "p0": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p0": "0.500", "p3": "0.125", "p1": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 871:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 5 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 06:15 to 16:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "10h30", "dur_3": "9h30", "dur_0": "10h15", "dur_1": "10h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "30", "answer_2": "24", "answer_0": "60", "answer_3": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, March 06, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1961. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "64", "age_2": "66", "age_3": "67", "age_0": "65"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "19", "answer_0": "33", "answer_3": "21", "answer_2": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = $651; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "6", "answer_1": "11", "answer_3": "15", "answer_0": "21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.700", "p2": "0.300", "p1": "0.300", "p3": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.167", "p0": "0.333", "p1": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.429", "p0": "0.571", "p2": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "28", "answer_2": "20", "answer_3": "8", "answer_1": "22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed May 05, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 12", "date_1": "May 10", "date_2": "May 10", "date_0": "May 12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.273", "p1": "0.727", "p3": "0.200", "p2": "0.727"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $345. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $210. A fee of $128 is charged, then a deposit of $49 arrives. A second fee of $64 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "71", "answer_2": "77", "answer_0": "67", "answer_3": "90"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $872. A fee of $181 is charged, then a deposit of $30 arrives. A second fee of $90 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "614", "answer_0": "631", "answer_1": "652", "answer_3": "650"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = 2026-08-26; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=868, unit_price=901, discount_pct=0, region=EU, sku_total=1769 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "782068", "v3": "781200", "v1": "1769", "v2": "78206"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $741. A fee of $43 is charged, then a deposit of $83 arrives. A second fee of $21 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "779", "answer_3": "765", "answer_0": "760", "answer_1": "738"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 20:45 to 21:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "0h45", "dur_2": "1h00", "dur_0": "0h30", "dur_3": "-1h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = 2026-08-21; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $488 per year. It increases by 15% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "414.8", "answer_1": "561.2", "answer_0": "542.2", "answer_3": "513.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Sunday, January 25, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = ops; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = ops; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $322. A fee of $119 is charged, then a deposit of $47 arrives. A second fee of $59 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "181", "answer_0": "191", "answer_1": "204", "answer_2": "174"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, May 25, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = $714; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=628, unit_price=364, discount_pct=0, region=EU, sku_total=992 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "228592", "v3": "227964", "v1": "992", "v2": "22859"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.333", "p3": "0.167", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.083", "p0": "0.250", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed February 13, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 24", "date_3": "February 22", "date_1": "February 20", "date_2": "February 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $605. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $608. A fee of $156 is charged, then a deposit of $44 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "418", "answer_3": "408", "answer_2": "445", "answer_1": "401"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 81 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "207", "answer_3": "243", "answer_0": "224", "answer_2": "210"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1984. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "43", "age_3": "44", "age_0": "42", "age_1": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=139, unit_price=702, discount_pct=0, region=EU, sku_total=841 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "9757", "v3": "97439", "v0": "97578", "v1": "841"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 16:15 to 22:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "5h30", "dur_1": "6h30", "dur_2": "6h30", "dur_0": "6h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 847:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, November 16, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_1": "16", "answer_0": "28", "answer_2": "14"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = $317; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 331:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $541 per year. It increases by 15% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "459.85", "answer_1": "622.15", "answer_3": "612.15", "answer_0": "599.15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $834. A fee of $151 is charged, then a deposit of $72 arrives. A second fee of $75 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "696", "answer_0": "680", "answer_1": "654", "answer_2": "650"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 378:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 234 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "84", "answer_2": "99", "answer_3": "90", "answer_1": "93"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "20", "answer_1": "24", "answer_2": "20", "answer_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $259. A fee of $147 is charged, then a deposit of $38 arrives. A second fee of $73 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "57", "answer_3": "57", "answer_1": "67", "answer_0": "77"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = pending; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 366 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "103", "answer_1": "133", "answer_3": "129", "answer_0": "120"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed April 19, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 26", "date_1": "April 24", "date_3": "April 26", "date_0": "April 24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed December 04, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 13", "date_0": "December 17", "date_3": "December 15", "date_2": "December 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.333", "p3": "0.625", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1982. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "44", "age_1": "42", "age_0": "43", "age_3": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $276. A fee of $72 is charged, then a deposit of $76 arrives. A second fee of $36 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "225", "answer_0": "244", "answer_2": "259", "answer_3": "214"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.750", "p2": "0.750", "p3": "0.143"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 381 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "336", "answer_1": "341", "answer_3": "342", "answer_2": "351"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1971. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "55", "age_1": "54", "age_2": "56", "age_3": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = $373; priority = low.\n\nClaim to verify: \"the status is $373\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.700", "p0": "0.727", "p1": "0.273", "p2": "0.273"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 4 grams of salt per serving and serves 10 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 133 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_2": "65", "answer_1": "55", "answer_3": "34"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 518:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = 2026-03-12; priority = high.\n\nClaim to verify: \"the date is 2026-03-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed March 10, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 18", "date_1": "March 16", "date_2": "March 20", "date_3": "March 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = 2026-05-22; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 11:00 to 14:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "3h30", "dur_3": "2h30", "dur_2": "3h30", "dur_0": "3h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 332 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_3": "111", "answer_1": "127", "answer_2": "133"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "40", "answer_1": "21", "answer_0": "56", "answer_2": "16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = ops; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, January 24, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1639 per year. It increases by 15% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1880.85", "answer_2": "1393.15", "answer_1": "1884.85", "answer_0": "1848.85"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1946 per year. It increases by 25% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2425.5", "answer_0": "2403.5", "answer_2": "1459.5", "answer_1": "2432.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.462", "p2": "0.462", "p3": "0.500", "p0": "0.538"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1837 per year. It increases by 20% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2168.4", "answer_3": "2151.4", "answer_1": "2204.4", "answer_2": "1469.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $797 per year. It increases by 10% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "717.3", "answer_1": "876.7", "answer_0": "843.7", "answer_3": "832.7"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=735, unit_price=309, discount_pct=0, region=EU, sku_total=1044 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "22711", "v3": "227850", "v1": "1044", "v0": "227115"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = $979; priority = low.\n\nClaim to verify: \"the amount is $979\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $738 per year. It increases by 25% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "886.5", "answer_2": "553.5", "answer_3": "874.5", "answer_1": "922.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed December 27, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 01", "date_3": "January 03", "date_0": "January 01", "date_2": "January 03"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 397 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "178", "answer_3": "180", "answer_2": "144", "answer_0": "160"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=650, unit_price=602, discount_pct=0, region=EU, sku_total=1252 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "391300", "v2": "39130", "v3": "391950", "v1": "1252"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = 2026-04-04; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 7 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 353 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "146", "answer_2": "158", "answer_0": "140", "answer_3": "150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 705:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed September 27, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 07", "date_0": "October 09", "date_2": "October 11", "date_3": "October 09"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $209. A fee of $153 is charged, then a deposit of $29 arrives. A second fee of $76 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "9", "answer_1": "-17", "answer_3": "6", "answer_2": "6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $17k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $12k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p1": "0.200", "p3": "0.050", "p0": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = $190; priority = low.\n\nClaim to verify: \"the status is $190\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, June 20, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = 2026-04-05; priority = low.\n\nClaim to verify: \"the owner is 2026-04-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = 2026-12-15; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $320. A fee of $86 is charged, then a deposit of $72 arrives. A second fee of $43 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "246", "answer_0": "263", "answer_2": "246", "answer_1": "254"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 133 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "204", "answer_0": "224", "answer_3": "215", "answer_1": "233"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $604. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_0": "48", "answer_2": "24", "answer_1": "28"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 158 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "126", "answer_2": "151", "answer_1": "146", "answer_0": "144"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "63", "answer_2": "14", "answer_1": "21", "answer_3": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.150", "p0": "0.200", "p1": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 454:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 809:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $263. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1987. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "36", "age_2": "38", "age_3": "35", "age_0": "37"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=743, unit_price=550, discount_pct=0, region=EU, sku_total=1293 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "40865", "v0": "408650", "v1": "1293", "v3": "407907"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed January 14, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 17", "date_3": "January 19", "date_2": "January 21", "date_0": "January 19"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.167", "p0": "0.333", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = $707; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = $102; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 20:30 to 22:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "1h00", "dur_1": "1h45", "dur_2": "1h15", "dur_3": "0h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 131 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "282", "answer_2": "272", "answer_3": "283", "answer_0": "270"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.667", "p0": "0.500", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.800", "p2": "0.200", "p1": "0.200", "p3": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Tuesday, November 10, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $628. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Saturday, February 28, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 191 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_2": "5", "answer_0": "12", "answer_1": "29"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p0": "0.250", "p3": "0.125", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 15:30 to 22:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "6h15", "dur_2": "6h45", "dur_3": "5h45", "dur_1": "6h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Friday, November 06, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.429", "p2": "0.533", "p0": "0.467", "p1": "0.533"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "48", "answer_1": "30", "answer_2": "24", "answer_0": "72"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $10k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1980. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "43", "age_3": "42", "age_0": "44", "age_2": "45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.615", "p3": "0.333", "p1": "0.615", "p0": "0.385"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=815, unit_price=673, discount_pct=0, region=EU, sku_total=1488 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "548495", "v1": "1488", "v3": "547680", "v2": "54849"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 741:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 06:00 to 12:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "6h45", "dur_1": "6h30", "dur_3": "5h30", "dur_0": "6h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 7 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 255 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "168", "answer_3": "163", "answer_1": "157", "answer_2": "163"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $242. A fee of $103 is charged, then a deposit of $83 arrives. A second fee of $51 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "156", "answer_1": "179", "answer_2": "175", "answer_0": "171"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "15", "answer_1": "17", "answer_0": "27", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = 2026-04-10; priority = high.\n\nClaim to verify: \"the status is 2026-04-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $411. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.167", "p1": "0.500", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p0": "0.250", "p2": "0.125", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_1": "19", "answer_2": "16", "answer_0": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 406:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, June 12, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 201 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "138", "answer_2": "110", "answer_3": "106", "answer_0": "120"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=936, unit_price=609, discount_pct=0, region=EU, sku_total=1545 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "570024", "v3": "570960", "v2": "57002", "v1": "1545"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = 2026-04-04; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1392 per year. It increases by 10% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1454.2", "answer_2": "1252.8", "answer_0": "1488.2", "answer_1": "1531.2"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1605 per year. It increases by 5% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1665.25", "answer_3": "1687.25", "answer_1": "1685.25", "answer_2": "1524.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 4 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 09:30 to 19:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "10h00", "dur_2": "9h45", "dur_0": "9h15", "dur_3": "9h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $228. A fee of $73 is charged, then a deposit of $81 arrives. A second fee of $36 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "184", "answer_0": "200", "answer_1": "189", "answer_2": "209"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 18:15 to 21:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "3h30", "dur_3": "2h30", "dur_2": "3h15", "dur_0": "3h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 6 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed January 18, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 26", "date_2": "January 26", "date_3": "January 28", "date_0": "January 28"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = $297; priority = low.\n\nClaim to verify: \"the amount is $297\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $843. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $827. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = $377; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1974. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "53", "age_3": "54", "age_0": "52", "age_1": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=640, unit_price=189, discount_pct=0, region=EU, sku_total=829 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "12096", "v0": "120960", "v1": "829", "v3": "120320"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1971. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "54", "age_0": "53", "age_1": "52", "age_3": "51"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "10", "answer_3": "10", "answer_1": "12", "answer_0": "20"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, November 17, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.333", "p0": "0.250", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "10", "answer_0": "20", "answer_1": "12", "answer_3": "10"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed December 19, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 28", "date_1": "December 27", "date_3": "December 29", "date_0": "December 30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed October 08, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 14", "date_0": "October 16", "date_3": "October 16", "date_2": "October 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.400", "p1": "0.500", "p0": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = 2026-02-23; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $302. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = 2026-02-28; priority = low.\n\nClaim to verify: \"the date is 2026-02-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 5 grams of salt per serving and serves 3 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $442 per year. It increases by 20% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "504.4", "answer_2": "353.6", "answer_1": "530.4", "answer_0": "488.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $784. A fee of $100 is charged, then a deposit of $76 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "691", "answer_2": "738", "answer_1": "698", "answer_0": "710"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_0": "44", "answer_2": "20", "answer_1": "26"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = 2026-05-16; priority = low.\n\nClaim to verify: \"the amount is 2026-05-16\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1975. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "50", "age_3": "49", "age_0": "51", "age_2": "52"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=454, unit_price=539, discount_pct=0, region=EU, sku_total=993 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "993", "v2": "24470", "v3": "244252", "v0": "244706"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, March 13, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = 2026-07-19; priority = high.\n\nClaim to verify: \"the amount is 2026-07-19\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.250", "p1": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = $487; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.083", "p0": "0.167", "p1": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1461 per year. It increases by 20% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1753.2", "answer_2": "1168.8", "answer_3": "1704.2", "answer_0": "1714.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.400", "p2": "0.500", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $722. A fee of $74 is charged, then a deposit of $24 arrives. A second fee of $37 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "635", "answer_2": "605", "answer_3": "605", "answer_1": "627"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.417", "p3": "0.364", "p2": "0.583", "p1": "0.583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.167", "p1": "0.833", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1970. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "56", "age_3": "57", "age_1": "54", "age_0": "55"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 12 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_0": "66", "answer_2": "24", "answer_1": "31"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 9 grams of salt per serving and serves 4 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.250", "p2": "0.167", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = closed; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed May 05, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 15", "date_3": "May 17", "date_0": "May 19", "date_2": "May 17"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.714", "p3": "0.667", "p1": "0.286", "p2": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, July 23, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_3": "12", "answer_0": "30", "answer_1": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 2 grams of salt per serving and serves 12 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = $605; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed January 15, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 26", "date_2": "January 26", "date_1": "January 24", "date_0": "January 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 8 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 323 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "390", "answer_0": "384", "answer_1": "365", "answer_3": "392"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1407 per year. It increases by 25% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1758.75", "answer_0": "1746.75", "answer_2": "1055.25", "answer_3": "1778.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, January 26, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_3": "12", "answer_1": "15", "answer_0": "24"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "91", "answer_1": "49", "answer_2": "42", "answer_3": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.200", "p2": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "33", "answer_3": "18", "answer_2": "15", "answer_1": "21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1962. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "60", "age_0": "62", "age_1": "61", "age_2": "63"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $236 per year. It increases by 15% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "200.6", "answer_3": "214.4", "answer_1": "271.4", "answer_0": "250.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $485. A fee of $125 is charged, then a deposit of $24 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "331", "answer_0": "322", "answer_3": "308", "answer_2": "309"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 10 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p1": "0.250", "p2": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.600", "p2": "0.400", "p3": "0.556", "p1": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $967 per year. It increases by 25% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1189.75", "answer_1": "1208.75", "answer_2": "725.25", "answer_3": "1171.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.467", "p3": "0.500", "p0": "0.533", "p1": "0.467"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Monday, September 28, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 646:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.778", "p0": "0.800", "p2": "0.200", "p1": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = $127; priority = high.\n\nClaim to verify: \"the amount is $127\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 7 grams of salt per serving and serves 4 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 162 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "90", "answer_3": "89", "answer_1": "110", "answer_0": "108"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1484 per year. It increases by 5% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1513.2", "answer_1": "1558.2", "answer_3": "1550.2", "answer_2": "1409.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.556", "p3": "0.375", "p2": "0.556", "p0": "0.444"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $286 per year. It increases by 15% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "243.1", "answer_1": "328.9", "answer_3": "313.9", "answer_0": "292.9"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 374 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "437", "answer_0": "432", "answer_3": "448", "answer_1": "447"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, December 09, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1983. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "42", "age_1": "40", "age_3": "43", "age_0": "41"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 726:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 3 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1974. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "50", "age_2": "51", "age_3": "52", "age_1": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=518, unit_price=840, discount_pct=0, region=EU, sku_total=1358 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "435638", "v2": "43512", "v0": "435120", "v1": "1358"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = $165; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: date = 2026-11-15; priority = low.\n\nClaim to verify: \"the date is 2026-11-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = $541; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $63. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $8k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed February 14, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 20", "date_2": "February 22", "date_3": "February 21", "date_1": "February 19"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_1": "22", "answer_2": "18", "answer_0": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $14k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed March 27, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 12", "date_0": "April 10", "date_3": "April 08", "date_1": "April 06"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed May 06, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 14", "date_1": "May 12", "date_0": "May 14", "date_2": "May 16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 760:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1961. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "63", "age_2": "64", "age_3": "61", "age_1": "62"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 3 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=763, unit_price=965, discount_pct=0, region=EU, sku_total=1728 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "736295", "v2": "73629", "v3": "737058", "v1": "1728"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $627 per year. It increases by 5% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "603.35", "answer_0": "614.35", "answer_2": "595.65", "answer_1": "658.35"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "35", "answer_2": "21", "answer_3": "14", "answer_1": "23"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=326, unit_price=979, discount_pct=0, region=EU, sku_total=1305 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "319154", "v1": "1305", "v3": "318828", "v2": "31915"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $316. A fee of $174 is charged, then a deposit of $36 arrives. A second fee of $87 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "77", "answer_0": "91", "answer_2": "106", "answer_1": "113"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1985. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "41", "age_1": "40", "age_2": "42", "age_3": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed February 19, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 22", "date_0": "February 24", "date_3": "February 24", "date_2": "February 26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p3": "0.125", "p2": "0.125", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $646 per year. It increases by 10% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "666.6", "answer_2": "581.4", "answer_3": "660.6", "answer_1": "710.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Friday, August 28, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1913 per year. It increases by 20% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2278.6", "answer_2": "1530.4", "answer_1": "2295.6", "answer_0": "2273.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Wednesday, March 18, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.200", "p0": "0.273", "p1": "0.727", "p2": "0.727"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = 2026-01-18; priority = high.\n\nClaim to verify: \"the amount is 2026-01-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 262:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed June 19, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 26", "date_3": "June 24", "date_0": "June 24", "date_1": "June 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_3": "24", "answer_1": "35", "answer_2": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=744, unit_price=304, discount_pct=0, region=EU, sku_total=1048 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "225432", "v0": "226176", "v1": "1048", "v2": "22617"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, March 20, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 14:15 to 17:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h45", "dur_0": "2h30", "dur_2": "3h00", "dur_3": "1h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=806, unit_price=701, discount_pct=0, region=EU, sku_total=1507 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "56500", "v1": "1507", "v3": "565812", "v0": "565006"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $603. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 217 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "280", "answer_2": "262", "answer_3": "267", "answer_1": "266"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p2": "0.375", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1959. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "66", "age_0": "67", "age_3": "65", "age_2": "68"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1988. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "37", "age_3": "39", "age_2": "38", "age_1": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.286", "p1": "0.625", "p2": "0.625", "p0": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = 2026-05-05; priority = low.\n\nClaim to verify: \"the date is 2026-05-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = $320; priority = high.\n\nClaim to verify: \"the amount is $320\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = 2026-02-23; priority = high.\n\nClaim to verify: \"the owner is 2026-02-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p0": "0.250", "p2": "0.125", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 06:15 to 17:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "9h45", "dur_1": "10h45", "dur_0": "10h00", "dur_2": "10h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 157 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "144", "answer_1": "161", "answer_3": "133", "answer_2": "137"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 376 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "177", "answer_0": "196", "answer_2": "179", "answer_3": "215"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = 2026-11-13; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_2": "21", "answer_0": "35", "answer_1": "23"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $286 per year. It increases by 25% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "357.5", "answer_3": "360.5", "answer_2": "214.5", "answer_0": "341.5"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=262, unit_price=242, discount_pct=0, region=EU, sku_total=504 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "63142", "v0": "63404", "v1": "504", "v2": "6340"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 5 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p1": "0.250", "p3": "0.050", "p2": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.429", "p3": "0.538", "p0": "0.571", "p1": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 848:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = open; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = 2026-01-18; priority = high.\n\nClaim to verify: \"the amount is 2026-01-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = $695; priority = high.\n\nClaim to verify: \"the amount is $695\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $701. A fee of $85 is charged, then a deposit of $45 arrives. A second fee of $42 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "615", "answer_0": "619", "answer_2": "633", "answer_3": "627"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_0": "32", "answer_3": "16", "answer_1": "18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=918, unit_price=380, discount_pct=0, region=EU, sku_total=1298 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "34884", "v3": "349758", "v1": "1298", "v0": "348840"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.300", "p2": "0.636", "p0": "0.364", "p1": "0.636"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 10 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 6 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 06:30 to 11:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "5h00", "dur_3": "4h00", "dur_2": "4h30", "dur_0": "4h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.455", "p2": "0.500", "p1": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = 2026-02-19; priority = high.\n\nClaim to verify: \"the amount is 2026-02-19\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, May 10, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p3": "0.167", "p0": "0.500", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.222", "p1": "0.222", "p3": "0.750", "p0": "0.778"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=478, unit_price=246, discount_pct=0, region=EU, sku_total=724 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "724", "v3": "118066", "v0": "117588", "v2": "11758"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=992, unit_price=291, discount_pct=0, region=EU, sku_total=1283 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "287680", "v0": "288672", "v2": "28867", "v1": "1283"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=456, unit_price=741, discount_pct=0, region=EU, sku_total=1197 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1197", "v2": "33789", "v3": "337440", "v0": "337896"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "9", "answer_0": "27", "answer_2": "18", "answer_1": "21"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 7 people. A shopper bought a 14 gram box and a 140 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 06:15 to 22:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "16h00", "dur_0": "15h45", "dur_2": "16h00", "dur_3": "15h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $545. A fee of $32 is charged, then a deposit of $20 arrives. A second fee of $16 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "515", "answer_2": "525", "answer_1": "536", "answer_0": "517"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $39k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = 2026-01-24; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 499:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = ops; priority = low.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=895, unit_price=378, discount_pct=0, region=EU, sku_total=1273 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "338310", "v1": "1273", "v2": "33831", "v3": "339205"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed July 25, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 31", "date_2": "July 29", "date_1": "July 30", "date_3": "August 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = $27; priority = low.\n\nClaim to verify: \"the owner is $27\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.700", "p1": "0.700", "p3": "0.222", "p0": "0.300"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.500", "p2": "0.333", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed October 03, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 15", "date_1": "October 10", "date_0": "October 13", "date_3": "October 12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed July 20, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 30", "date_0": "July 28", "date_3": "July 28", "date_1": "July 26"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $369. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=123, unit_price=946, discount_pct=0, region=EU, sku_total=1069 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "116358", "v3": "116481", "v2": "11635", "v1": "1069"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.538", "p1": "0.538", "p0": "0.462", "p3": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=810, unit_price=268, discount_pct=0, region=EU, sku_total=1078 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1078", "v2": "21708", "v0": "217080", "v3": "216270"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_3": "16", "answer_2": "32", "answer_1": "34"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=928, unit_price=393, discount_pct=0, region=EU, sku_total=1321 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "36470", "v1": "1321", "v0": "364704", "v3": "363776"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 850:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $719. A fee of $105 is charged, then a deposit of $28 arrives. A second fee of $52 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "614", "answer_2": "586", "answer_1": "601", "answer_0": "590"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 218 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "87", "answer_3": "100", "answer_0": "96", "answer_2": "112"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = 2026-12-26; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $35k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $25k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: owner = 2026-11-21; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = $619; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "17", "answer_2": "10", "answer_0": "45", "answer_3": "35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 832:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $13k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 235 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_1": "36", "answer_3": "24", "answer_2": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.533", "p0": "0.467", "p3": "0.429", "p1": "0.533"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "9", "answer_0": "30", "answer_3": "21", "answer_1": "16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1977. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "47", "age_2": "49", "age_0": "48", "age_3": "50"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 7 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "30", "answer_2": "18", "answer_3": "12", "answer_1": "22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.333", "p3": "0.167", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed November 14, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 19", "date_2": "November 21", "date_3": "November 20", "date_1": "November 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = $863; priority = high.\n\nClaim to verify: \"the status is $863\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 510:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1970. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "53", "age_2": "55", "age_3": "56", "age_0": "54"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = $884; priority = high.\n\nClaim to verify: \"the amount is $884\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = $229; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 208 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "131", "answer_3": "145", "answer_2": "143", "answer_0": "150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed April 06, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 11", "date_3": "April 11", "date_0": "April 09", "date_1": "April 09"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1058 per year. It increases by 5% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1110.9", "answer_3": "1046.9", "answer_0": "1066.9", "answer_2": "1005.1"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 734:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=953, unit_price=408, discount_pct=0, region=EU, sku_total=1361 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "389777", "v2": "38882", "v0": "388824", "v1": "1361"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1982. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "41", "age_0": "42", "age_3": "44", "age_2": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $377. A fee of $58 is charged, then a deposit of $56 arrives. A second fee of $29 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "351", "answer_0": "346", "answer_1": "368", "answer_3": "333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.400", "p0": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = $911; priority = low.\n\nClaim to verify: \"the amount is $911\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1963. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "64", "age_0": "63", "age_3": "61", "age_1": "62"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.273", "p0": "0.727", "p1": "0.273", "p3": "0.700"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $722. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "26", "answer_2": "21", "answer_0": "56", "answer_3": "35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $847. A fee of $114 is charged, then a deposit of $82 arrives. A second fee of $57 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "758", "answer_1": "764", "answer_3": "777", "answer_2": "771"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p3": "0.050", "p1": "0.200", "p0": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p0": "0.500", "p3": "0.429"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 7 grams of salt per serving and serves 12 people. A shopper bought a 84 gram box and a 840 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $541 per year. It increases by 25% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "693.25", "answer_2": "405.75", "answer_0": "656.25", "answer_1": "676.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 6 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=504, unit_price=378, discount_pct=0, region=EU, sku_total=882 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "19051", "v3": "191016", "v0": "190512", "v1": "882"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 853:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed February 14, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 18", "date_0": "February 19", "date_3": "February 20", "date_2": "February 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 378:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: date = $199; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $451 per year. It increases by 5% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "392.55", "answer_0": "428.55", "answer_1": "473.55", "answer_2": "428.45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = 2026-07-09; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 6 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=673, unit_price=323, discount_pct=0, region=EU, sku_total=996 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "21737", "v1": "996", "v3": "218052", "v0": "217379"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed May 18, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 27", "date_3": "May 27", "date_1": "May 25", "date_2": "May 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $490. A fee of $94 is charged, then a deposit of $51 arrives. A second fee of $47 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "380", "answer_1": "382", "answer_2": "407", "answer_0": "400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 702:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_1": "14", "answer_0": "20", "answer_3": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "6", "answer_0": "27", "answer_1": "13", "answer_3": "21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.500", "p2": "0.500", "p3": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed November 15, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 19", "date_2": "November 17", "date_0": "November 19", "date_3": "November 21"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.333", "p3": "0.625", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: amount = $688; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = 2026-03-27; priority = low.\n\nClaim to verify: \"the status is 2026-03-27\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 316 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "13", "answer_3": "22", "answer_0": "32", "answer_1": "37"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = 2026-03-04; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p0": "0.800", "p2": "0.200", "p3": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed September 16, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 26", "date_0": "September 30", "date_3": "September 28", "date_2": "September 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1980. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "45", "age_0": "44", "age_3": "46", "age_1": "43"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 626:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=253, unit_price=607, discount_pct=0, region=EU, sku_total=860 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "153571", "v1": "860", "v3": "153824", "v2": "15357"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 152 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "16", "answer_0": "12", "answer_2": "22", "answer_3": "14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1906 per year. It increases by 5% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1956.3", "answer_2": "1810.7", "answer_0": "1963.3", "answer_1": "2001.3"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 52 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "240", "answer_1": "222", "answer_2": "228", "answer_3": "259"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "9", "answer_2": "18", "answer_0": "27", "answer_1": "21"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, July 06, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_1": "24", "answer_3": "20", "answer_2": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.273", "p3": "0.700", "p0": "0.727", "p1": "0.273"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 9 grams of salt per serving and serves 11 people. A shopper bought a 99 gram box and a 990 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Thursday, January 22, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "42", "answer_3": "30", "answer_2": "12", "answer_1": "17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $486. A fee of $73 is charged, then a deposit of $57 arrives. A second fee of $36 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "434", "answer_3": "429", "answer_2": "447", "answer_1": "422"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=521, unit_price=923, discount_pct=0, region=EU, sku_total=1444 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "48088", "v1": "1444", "v3": "481404", "v0": "480883"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 815:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 743:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = $796; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $453. A fee of $57 is charged, then a deposit of $31 arrives. A second fee of $28 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "371", "answer_0": "399", "answer_2": "405", "answer_1": "377"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=658, unit_price=360, discount_pct=0, region=EU, sku_total=1018 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "23688", "v1": "1018", "v0": "236880", "v3": "236222"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p0": "0.150", "p3": "0.050", "p1": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $169. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, January 12, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $664. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $24k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1989. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "35", "age_2": "36", "age_1": "34", "age_3": "37"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "14", "answer_0": "18", "answer_3": "6", "answer_2": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 179 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "81", "answer_1": "76", "answer_2": "73", "answer_0": "84"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed January 04, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 16", "date_3": "January 16", "date_1": "January 14", "date_2": "January 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = $912; priority = high.\n\nClaim to verify: \"the date is $912\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = open; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 500:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 08:30 to 16:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "7h00", "dur_1": "8h00", "dur_2": "8h15", "dur_0": "7h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = $584; priority = high.\n\nClaim to verify: \"the amount is $584\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1976. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "49", "age_2": "51", "age_3": "52", "age_0": "50"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "6", "answer_2": "12", "answer_1": "14", "answer_0": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "21", "answer_3": "20", "answer_0": "36", "answer_2": "16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, June 22, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = ops; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1324 per year. It increases by 10% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1425.4", "answer_2": "1191.6", "answer_1": "1456.4", "answer_0": "1408.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 718:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1981. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "41", "age_2": "44", "age_0": "43", "age_1": "42"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Tuesday, December 08, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 09:00 to 20:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "10h30", "dur_2": "11h15", "dur_0": "10h45", "dur_1": "11h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = 2026-01-26; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = $323; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p2": "0.083", "p3": "0.083", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $577 per year. It increases by 10% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "519.3", "answer_1": "634.7", "answer_3": "566.7", "answer_0": "589.7"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1299 per year. It increases by 10% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1385.9", "answer_2": "1169.1", "answer_3": "1351.9", "answer_1": "1428.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "28", "answer_0": "70", "answer_2": "42", "answer_1": "46"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=146, unit_price=919, discount_pct=0, region=EU, sku_total=1065 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "13417", "v3": "134320", "v1": "1065", "v0": "134174"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 626:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 282 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "225", "answer_2": "220", "answer_3": "216", "answer_1": "210"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Friday, June 26, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $277. A fee of $70 is charged, then a deposit of $46 arrives. A second fee of $35 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "222", "answer_0": "218", "answer_3": "242", "answer_1": "209"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.700", "p2": "0.273", "p1": "0.273", "p0": "0.727"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = $926; priority = low.\n\nClaim to verify: \"the amount is $926\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed March 22, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 27", "date_3": "March 29", "date_1": "March 27", "date_2": "March 25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $444. A fee of $190 is charged, then a deposit of $59 arrives. A second fee of $95 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "203", "answer_2": "202", "answer_0": "218", "answer_3": "247"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.833", "p3": "0.167", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = 2026-10-07; priority = low.\n\nClaim to verify: \"the date is 2026-10-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.714", "p3": "0.667", "p2": "0.286", "p1": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p2": "0.150", "p1": "0.250", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.250", "p2": "0.150", "p0": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p2": "0.600", "p0": "0.400", "p3": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $498. A fee of $85 is charged, then a deposit of $41 arrives. A second fee of $42 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "425", "answer_3": "439", "answer_2": "419", "answer_0": "412"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 357 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "45", "answer_3": "53", "answer_1": "42", "answer_2": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=263, unit_price=751, discount_pct=0, region=EU, sku_total=1014 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "197513", "v1": "1014", "v3": "197250", "v2": "19751"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = ops; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "8", "answer_0": "28", "answer_2": "20", "answer_1": "22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = 2026-11-02; priority = high.\n\nClaim to verify: \"the status is 2026-11-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, July 16, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1964. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "60", "age_3": "63", "age_2": "62", "age_0": "61"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 195 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "78", "answer_2": "84", "answer_0": "98", "answer_3": "115"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 7 people. A shopper bought a 14 gram box and a 140 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed July 27, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "August 06", "date_2": "August 08", "date_1": "August 04", "date_3": "August 06"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $871 per year. It increases by 5% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "867.55", "answer_3": "843.55", "answer_2": "827.45", "answer_1": "914.55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1712 per year. It increases by 10% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1837.2", "answer_3": "1821.2", "answer_1": "1883.2", "answer_2": "1540.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.500", "p1": "0.833", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = 2026-07-25; priority = high.\n\nClaim to verify: \"the owner is 2026-07-25\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1990. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "36", "age_1": "35", "age_2": "37", "age_3": "34"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $589. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 8 grams of salt per serving and serves 11 people. A shopper bought a 88 gram box and a 880 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1969. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "55", "age_0": "56", "age_3": "58", "age_2": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.583", "p0": "0.417", "p2": "0.583", "p3": "0.364"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 762:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1987. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "39", "age_1": "38", "age_3": "37", "age_2": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = $67; priority = low.\n\nClaim to verify: \"the owner is $67\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $759. A fee of $123 is charged, then a deposit of $47 arrives. A second fee of $61 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "594", "answer_3": "614", "answer_2": "635", "answer_0": "622"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.533", "p1": "0.467", "p2": "0.467"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 289 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "25", "answer_1": "59", "answer_0": "42", "answer_2": "36"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 5 grams of salt per serving and serves 8 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 306 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_2": "40", "answer_3": "40", "answer_1": "79"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.333", "p0": "0.667", "p3": "0.636"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 393 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_0": "36", "answer_3": "54", "answer_1": "53"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $972 per year. It increases by 10% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1035.2", "answer_1": "1069.2", "answer_3": "1058.2", "answer_2": "874.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, October 24, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.462", "p1": "0.538", "p2": "0.538", "p3": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1973. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "49", "age_0": "51", "age_1": "50", "age_2": "52"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1972. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "54", "age_0": "53", "age_3": "51", "age_1": "52"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=997, unit_price=662, discount_pct=0, region=EU, sku_total=1659 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "659017", "v1": "1659", "v2": "66001", "v0": "660014"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 473:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = $441; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1971. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "55", "age_3": "52", "age_0": "54", "age_1": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=967, unit_price=442, discount_pct=0, region=EU, sku_total=1409 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1409", "v0": "427414", "v2": "42741", "v3": "428381"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=272, unit_price=822, discount_pct=0, region=EU, sku_total=1094 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "22358", "v3": "223856", "v1": "1094", "v0": "223584"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "9", "answer_2": "6", "answer_0": "15", "answer_1": "9"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1055 per year. It increases by 5% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1106.75", "answer_2": "1002.25", "answer_1": "1107.75", "answer_0": "1082.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 185 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "103", "answer_1": "120", "answer_0": "112", "answer_3": "110"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_3": "24", "answer_0": "40", "answer_1": "19"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $540. A fee of $79 is charged, then a deposit of $22 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "449", "answer_3": "459", "answer_0": "444", "answer_2": "422"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "39", "answer_3": "21", "answer_2": "18", "answer_1": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, May 24, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 772:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 5 grams of salt per serving and serves 9 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "32", "answer_3": "35", "answer_0": "60", "answer_2": "25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.500", "p0": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p2": "0.375", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed March 15, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 21", "date_0": "March 19", "date_1": "March 19", "date_2": "March 21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed March 27, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 03", "date_3": "April 03", "date_2": "April 01", "date_1": "April 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1970. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "55", "age_2": "56", "age_3": "57", "age_1": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 335 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "329", "answer_2": "333", "answer_0": "324", "answer_1": "316"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 195 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "128", "answer_0": "135", "answer_2": "154", "answer_3": "145"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 849:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.500", "p0": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 06:15 to 18:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "11h15", "dur_2": "11h45", "dur_3": "10h45", "dur_1": "11h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 952:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $480. A fee of $180 is charged, then a deposit of $51 arrives. A second fee of $90 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "243", "answer_1": "249", "answer_2": "240", "answer_0": "261"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, December 15, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 3 people. A shopper bought a 9 gram box and a 90 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: date = $722; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_3": "42", "answer_1": "25", "answer_2": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = 2026-05-04; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p3": "0.050", "p1": "0.200", "p0": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 83 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "125", "answer_3": "112", "answer_1": "122", "answer_0": "108"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = $268; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.667", "p2": "0.500", "p1": "0.833"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 12 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 601:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.429", "p0": "0.571", "p3": "0.500", "p1": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $707. A fee of $51 is charged, then a deposit of $34 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "686", "answer_3": "641", "answer_0": "665", "answer_1": "687"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 194 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "71", "answer_2": "78", "answer_0": "63", "answer_1": "55"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=399, unit_price=270, discount_pct=0, region=EU, sku_total=669 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "669", "v2": "10773", "v0": "107730", "v3": "108129"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = 2026-05-15; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=334, unit_price=222, discount_pct=0, region=EU, sku_total=556 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "7414", "v1": "556", "v0": "74148", "v3": "73814"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = $248; priority = low.\n\nClaim to verify: \"the status is $248\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 7 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = 2026-05-11; priority = high.\n\nClaim to verify: \"the status is 2026-05-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $263. A fee of $136 is charged, then a deposit of $41 arrives. A second fee of $68 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "100", "answer_2": "123", "answer_3": "126", "answer_1": "122"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $2000 per year. It increases by 10% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2168", "answer_3": "2137", "answer_2": "1800", "answer_1": "2200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=880, unit_price=746, discount_pct=0, region=EU, sku_total=1626 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "655600", "v0": "656480", "v2": "65648", "v1": "1626"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p2": "0.083", "p3": "0.083", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p1": "0.375", "p3": "0.125", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 5 people. A shopper bought a 10 gram box and a 100 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 569:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $38k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=913, unit_price=985, discount_pct=0, region=EU, sku_total=1898 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "899305", "v1": "1898", "v3": "900218", "v2": "89930"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, May 12, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = 2026-08-12; priority = low.\n\nClaim to verify: \"the date is 2026-08-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $236. A fee of $162 is charged, then a deposit of $23 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "7", "answer_2": "45", "answer_3": "12", "answer_0": "16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $568 per year. It increases by 20% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "640.6", "answer_3": "664.6", "answer_1": "681.6", "answer_2": "454.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.286", "p0": "0.714", "p1": "0.286", "p3": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1968 per year. It increases by 25% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1476", "answer_0": "2431", "answer_3": "2440", "answer_1": "2460"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=478, unit_price=295, discount_pct=0, region=EU, sku_total=773 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "773", "v0": "141010", "v3": "140532", "v2": "14101"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.533", "p0": "0.467", "p1": "0.533", "p3": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1977 per year. It increases by 5% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2075.85", "answer_2": "1878.15", "answer_3": "2033.85", "answer_0": "2047.85"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.625", "p2": "0.375", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "20", "answer_3": "25", "answer_2": "15", "answer_0": "40"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 12 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 312 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "110", "answer_0": "120", "answer_3": "125", "answer_2": "117"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed December 21, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 29", "date_3": "December 29", "date_2": "December 27", "date_1": "December 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "32", "answer_1": "35", "answer_3": "24", "answer_0": "56"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=581, unit_price=595, discount_pct=0, region=EU, sku_total=1176 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "345695", "v3": "345114", "v2": "34569", "v1": "1176"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 142 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "175", "answer_2": "167", "answer_0": "180", "answer_3": "184"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = $203; priority = high.\n\nClaim to verify: \"the owner is $203\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.083", "p0": "0.167", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = $883; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p2": "0.050", "p3": "0.050", "p1": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1978. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "48", "age_0": "46", "age_1": "45", "age_2": "47"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $314. A fee of $78 is charged, then a deposit of $26 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "211", "answer_0": "223", "answer_2": "243", "answer_3": "194"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 538:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $797 per year. It increases by 10% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "827.7", "answer_0": "865.7", "answer_1": "876.7", "answer_2": "717.3"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.400", "p3": "0.500", "p2": "0.400", "p0": "0.600"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1986. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "40", "age_3": "41", "age_1": "38", "age_0": "39"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 9 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed April 03, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 10", "date_0": "April 10", "date_2": "April 08", "date_1": "April 08"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, June 02, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1983. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "42", "age_1": "40", "age_3": "43", "age_0": "41"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Thursday, August 13, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 14:15 to 15:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "1h00", "dur_0": "0h45", "dur_3": "0h00", "dur_2": "1h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $476 per year. It increases by 20% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "571.2", "answer_0": "539.2", "answer_2": "380.8", "answer_3": "560.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = 2026-02-22; priority = high.\n\nClaim to verify: \"the amount is 2026-02-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_3": "16", "answer_1": "20", "answer_0": "32"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $635. A fee of $62 is charged, then a deposit of $48 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "618", "answer_3": "570", "answer_0": "590", "answer_1": "597"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $2k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p1": "0.250", "p3": "0.083", "p2": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 9 grams of salt per serving and serves 11 people. A shopper bought a 99 gram box and a 990 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = 2026-11-03; priority = high.\n\nClaim to verify: \"the date is 2026-11-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $675. A fee of $35 is charged, then a deposit of $31 arrives. A second fee of $17 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "626", "answer_2": "638", "answer_1": "665", "answer_0": "654"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.545", "p2": "0.417", "p0": "0.583", "p1": "0.417"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 7 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1373 per year. It increases by 15% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1531.95", "answer_3": "1563.95", "answer_1": "1578.95", "answer_2": "1167.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 6 grams of salt per serving and serves 12 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Friday, January 23, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 7 people. A shopper bought a 49 gram box and a 490 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed January 14, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 20", "date_3": "January 20", "date_1": "January 18", "date_2": "January 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 5 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = $372; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p0": "0.250", "p1": "0.375", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 07:45 to 10:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "2h45", "dur_1": "2h30", "dur_3": "1h30", "dur_0": "2h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=491, unit_price=157, discount_pct=0, region=EU, sku_total=648 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "77578", "v0": "77087", "v1": "648", "v2": "7708"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 5 grams of salt per serving and serves 12 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1976. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "49", "age_2": "51", "age_3": "52", "age_0": "50"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.462", "p3": "0.500", "p2": "0.462", "p0": "0.538"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 230 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "360", "answer_3": "352", "answer_2": "341", "answer_1": "350"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=467, unit_price=340, discount_pct=0, region=EU, sku_total=807 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "159247", "v0": "158780", "v2": "15878", "v1": "807"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = $869; priority = low.\n\nClaim to verify: \"the amount is $869\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1884 per year. It increases by 10% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2072.4", "answer_0": "2057.4", "answer_3": "2071.4", "answer_2": "1695.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.615", "p3": "0.333", "p0": "0.385", "p2": "0.615"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed December 25, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 02", "date_3": "December 31", "date_1": "December 29", "date_0": "December 31"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, February 19, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 8 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p3": "0.167", "p2": "0.333", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = 2026-01-04; priority = low.\n\nClaim to verify: \"the date is 2026-01-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $499. A fee of $179 is charged, then a deposit of $37 arrives. A second fee of $89 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "260", "answer_1": "246", "answer_3": "252", "answer_0": "268"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed June 11, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 15", "date_0": "June 17", "date_3": "June 17", "date_2": "June 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = $130; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.125", "p1": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed March 07, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 14", "date_3": "March 16", "date_0": "March 17", "date_2": "March 19"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.500", "p0": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 74 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "106", "answer_1": "134", "answer_0": "126", "answer_3": "107"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1958. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "66", "age_3": "65", "age_0": "67", "age_2": "68"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.500", "p2": "0.375", "p1": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=258, unit_price=901, discount_pct=0, region=EU, sku_total=1159 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "232458", "v2": "23245", "v3": "232716", "v1": "1159"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = 2026-11-02; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 361:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = $968; priority = high.\n\nClaim to verify: \"the date is $968\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "36", "answer_1": "42", "answer_0": "72", "answer_3": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1988. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "40", "age_0": "38", "age_1": "37", "age_2": "39"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "63", "age_0": "64", "age_2": "65", "age_3": "62"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, February 24, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 302 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "129", "answer_2": "93", "answer_3": "95", "answer_0": "112"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.583", "p1": "0.417", "p2": "0.417", "p3": "0.545"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 51 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "313", "answer_2": "298", "answer_3": "310", "answer_0": "315"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed December 18, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 26", "date_0": "December 28", "date_1": "December 24", "date_2": "December 30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1008 per year. It increases by 25% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1190", "answer_0": "1212", "answer_2": "756", "answer_1": "1260"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.250", "p0": "0.167", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "10", "answer_1": "12", "answer_3": "10", "answer_0": "20"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $842. A fee of $106 is charged, then a deposit of $76 arrives. A second fee of $53 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "759", "answer_3": "753", "answer_2": "777", "answer_1": "741"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 691:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = $989; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.833", "p0": "0.667", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1988. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "37", "age_2": "39", "age_3": "40", "age_0": "38"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1988. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "38", "age_1": "37", "age_2": "39", "age_3": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = 2026-01-09; priority = high.\n\nClaim to verify: \"the amount is 2026-01-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p2": "0.100", "p3": "0.050", "p0": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = $595; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.083", "p2": "0.250", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = $652; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = 2026-02-03; priority = high.\n\nClaim to verify: \"the owner is 2026-02-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 11 people. A shopper bought a 66 gram box and a 660 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "31", "answer_0": "80", "answer_2": "24", "answer_3": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.200", "p2": "0.150", "p1": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An account starts at $729. A fee of $63 is charged, then a deposit of $56 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "697", "answer_3": "715", "answer_0": "691", "answer_1": "673"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p3": "0.636", "p1": "0.333", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Wednesday, August 12, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "78", "answer_3": "42", "answer_1": "43", "answer_2": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = 2026-09-03; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 597:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, April 02, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "16", "answer_1": "16", "answer_0": "28", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = $896; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 760:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $478. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $906; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 5 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $565. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $286 per year. It increases by 5% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "259.3", "answer_2": "271.7", "answer_1": "300.3", "answer_3": "252.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = $34; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, March 26, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = open; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: status = closed; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 17:45 to 22:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "4h30", "dur_3": "3h45", "dur_1": "4h45", "dur_2": "5h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 250:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, September 24, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $718. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $253. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=513, unit_price=586, discount_pct=0, region=EU, sku_total=1099 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "301131", "v1": "1099", "v2": "30061", "v0": "300618"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = 2026-09-23; priority = low.\n\nClaim to verify: \"the owner is 2026-09-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $645 per year. It increases by 15% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "741.75", "answer_0": "706.75", "answer_2": "548.25", "answer_3": "678.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 18:30 to 20:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h00", "dur_0": "1h45", "dur_3": "1h00", "dur_2": "2h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1433 per year. It increases by 25% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1818.25", "answer_2": "1074.75", "answer_0": "1778.25", "answer_1": "1791.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $403. A fee of $156 is charged, then a deposit of $75 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "266", "answer_0": "244", "answer_2": "221", "answer_1": "246"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = 2026-10-11; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 07:00 to 18:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "11h00", "dur_3": "10h15", "dur_1": "11h15", "dur_2": "11h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 317 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "83", "answer_3": "80", "answer_0": "72", "answer_2": "84"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $25k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "52", "answer_2": "48", "answer_0": "80", "answer_3": "32"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=953, unit_price=311, discount_pct=0, region=EU, sku_total=1264 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1264", "v3": "295430", "v2": "29638", "v0": "296383"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_1": "33", "answer_2": "28", "answer_0": "63"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = $380; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_2": "12", "answer_1": "16", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $334. A fee of $190 is charged, then a deposit of $65 arrives. A second fee of $95 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "116", "answer_1": "100", "answer_0": "114", "answer_2": "134"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1990. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "34", "age_2": "35", "age_1": "33", "age_3": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_3": "24", "answer_0": "48", "answer_1": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = $784; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = $906; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.222", "p0": "0.778", "p3": "0.750", "p2": "0.222"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1558 per year. It increases by 10% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1648.8", "answer_0": "1687.8", "answer_2": "1402.2", "answer_1": "1713.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=187, unit_price=993, discount_pct=0, region=EU, sku_total=1180 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "18569", "v1": "1180", "v0": "185691", "v3": "185504"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, December 03, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 63 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "264", "answer_0": "245", "answer_3": "260", "answer_2": "241"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 536:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, November 20, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.400", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.500", "p3": "0.444", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1959. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "65", "age_2": "66", "age_1": "64", "age_3": "63"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p3": "0.125", "p1": "0.375", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=172, unit_price=500, discount_pct=0, region=EU, sku_total=672 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "85828", "v0": "86000", "v1": "672", "v2": "8600"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.600", "p2": "0.333", "p1": "0.333", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1989. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "35", "age_2": "37", "age_3": "34", "age_0": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = 2026-11-17; priority = high.\n\nClaim to verify: \"the owner is 2026-11-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.083", "p2": "0.167", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 831:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 15:45 to 19:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "3h15", "dur_1": "3h30", "dur_3": "2h30", "dur_2": "3h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 10:45 to 17:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "7h00", "dur_2": "7h15", "dur_0": "6h45", "dur_3": "6h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.375", "p3": "0.125", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $157. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.333", "p0": "0.250", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 207:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 398 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "30", "answer_3": "13", "answer_1": "35", "answer_2": "14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_2": "18", "answer_1": "21", "answer_3": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1966. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "56", "age_0": "58", "age_2": "59", "age_1": "57"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = $29; priority = low.\n\nClaim to verify: \"the status is $29\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.429", "p2": "0.500", "p0": "0.500", "p1": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1961. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "64", "age_2": "66", "age_0": "65", "age_3": "67"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $952 per year. It increases by 25% this year, then a $40 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1150", "answer_3": "1128", "answer_2": "714", "answer_1": "1190"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=482, unit_price=237, discount_pct=0, region=EU, sku_total=719 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "11423", "v3": "113752", "v0": "114234", "v1": "719"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 3 grams of salt per serving and serves 11 people. A shopper bought a 33 gram box and a 330 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = 2026-11-24; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.636", "p1": "0.333", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "20", "answer_1": "23", "answer_0": "32", "answer_3": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Thursday, July 02, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.250", "p1": "0.417", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 234:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = open; priority = high.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=646, unit_price=996, discount_pct=0, region=EU, sku_total=1642 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "64341", "v3": "644062", "v0": "643416", "v1": "1642"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $679. A fee of $52 is charged, then a deposit of $23 arrives. A second fee of $26 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "604", "answer_1": "630", "answer_2": "639", "answer_0": "624"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = 2026-03-15; priority = low.\n\nClaim to verify: \"the amount is 2026-03-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $606. A fee of $57 is charged, then a deposit of $59 arrives. A second fee of $28 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "589", "answer_2": "559", "answer_0": "580", "answer_1": "591"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=470, unit_price=753, discount_pct=0, region=EU, sku_total=1223 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "354380", "v0": "353910", "v2": "35391", "v1": "1223"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=833, unit_price=249, discount_pct=0, region=EU, sku_total=1082 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1082", "v2": "20741", "v0": "207417", "v3": "206584"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, April 10, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=554, unit_price=759, discount_pct=0, region=EU, sku_total=1313 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "42048", "v3": "421040", "v0": "420486", "v1": "1313"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.417", "p3": "0.364", "p1": "0.583", "p2": "0.583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = ops; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = $812; priority = low.\n\nClaim to verify: \"the amount is $812\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=549, unit_price=871, discount_pct=0, region=EU, sku_total=1420 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "478179", "v1": "1420", "v2": "47817", "v3": "478728"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = 2026-09-04; priority = low.\n\nClaim to verify: \"the owner is 2026-09-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 489:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 218 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "150", "answer_3": "155", "answer_1": "165", "answer_0": "168"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = 2026-02-03; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 9 grams of salt per serving and serves 4 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $592. A fee of $153 is charged, then a deposit of $45 arrives. A second fee of $76 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "408", "answer_3": "412", "answer_1": "383", "answer_2": "399"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 612:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 67 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "69", "answer_0": "60", "answer_2": "54", "answer_1": "50"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $9k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $660. A fee of $151 is charged, then a deposit of $70 arrives. A second fee of $75 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "504", "answer_1": "516", "answer_2": "495", "answer_3": "490"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed October 15, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 19", "date_2": "October 23", "date_0": "October 21", "date_3": "October 21"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1976. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "49", "age_0": "48", "age_3": "46", "age_1": "47"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = 2026-09-15; priority = low.\n\nClaim to verify: \"the date is 2026-09-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "37", "answer_3": "42", "answer_2": "30", "answer_0": "72"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "8", "answer_0": "24", "answer_3": "16", "answer_1": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = $724; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=571, unit_price=295, discount_pct=0, region=EU, sku_total=866 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "866", "v0": "168445", "v2": "16844", "v3": "169016"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=806, unit_price=896, discount_pct=0, region=EU, sku_total=1702 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "72217", "v3": "722982", "v0": "722176", "v1": "1702"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1911 per year. It increases by 25% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1433.25", "answer_1": "2388.75", "answer_3": "2359.75", "answer_0": "2351.75"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 09:45 to 15:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "5h30", "dur_1": "5h45", "dur_3": "4h45", "dur_2": "5h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = 2026-06-28; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.462", "p2": "0.500", "p0": "0.500", "p1": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = pending; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $791. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.429", "p2": "0.429", "p0": "0.571", "p3": "0.538"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p3": "0.050", "p1": "0.200", "p0": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 878:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Wednesday, October 14, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=788, unit_price=753, discount_pct=0, region=EU, sku_total=1541 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "592576", "v0": "593364", "v2": "59336", "v1": "1541"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = open; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 524:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "6", "answer_0": "15", "answer_3": "9", "answer_1": "9"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 368:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 629:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=144, unit_price=762, discount_pct=0, region=EU, sku_total=906 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "906", "v2": "10972", "v0": "109728", "v3": "109584"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = 2026-12-22; priority = high.\n\nClaim to verify: \"the status is 2026-12-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $613. A fee of $44 is charged, then a deposit of $43 arrives. A second fee of $22 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "575", "answer_2": "604", "answer_0": "590", "answer_1": "606"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, October 24, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $728 per year. It increases by 15% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "618.8", "answer_1": "837.2", "answer_0": "821.2", "answer_3": "812.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p0": "0.100", "p3": "0.050", "p1": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Form fields: status = $727; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "34", "answer_3": "24", "answer_0": "54", "answer_2": "30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.333", "p2": "0.167", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1601 per year. It increases by 20% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1280.8", "answer_3": "1881.2", "answer_1": "1921.2", "answer_0": "1897.2"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.429", "p0": "0.571", "p3": "0.538", "p2": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p2": "0.100", "p0": "0.150", "p3": "0.050"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 07:00 to 07:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h15", "dur_1": "0h15", "dur_2": "0h00", "dur_0": "-1h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $787. A fee of $112 is charged, then a deposit of $54 arrives. A second fee of $56 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "686", "answer_3": "684", "answer_1": "656", "answer_0": "673"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $280. A fee of $168 is charged, then a deposit of $41 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "74", "answer_2": "89", "answer_0": "69", "answer_1": "88"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 6 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed May 25, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 27", "date_3": "May 31", "date_1": "May 29", "date_0": "May 29"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p3": "0.125", "p2": "0.125", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $394 per year. It increases by 10% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "431.4", "answer_2": "354.6", "answer_1": "433.4", "answer_0": "410.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 13:00 to 17:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "4h15", "dur_1": "4h45", "dur_2": "4h30", "dur_3": "3h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1984. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "42", "age_3": "40", "age_2": "43", "age_1": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.167", "p2": "0.167", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.333", "p1": "0.417", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1977. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "47", "age_2": "48", "age_1": "46", "age_3": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.455", "p0": "0.500", "p2": "0.500", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed September 26, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 29", "date_0": "October 01", "date_1": "September 30", "date_3": "October 02"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.200", "p2": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 179 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "68", "answer_3": "77", "answer_2": "66", "answer_0": "72"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 3 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 709:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.700", "p2": "0.700", "p3": "0.222", "p0": "0.300"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = 2026-10-22; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1975. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "49", "age_2": "50", "age_3": "47", "age_1": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=372, unit_price=150, discount_pct=0, region=EU, sku_total=522 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "55800", "v2": "5580", "v1": "522", "v3": "56172"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $219. A fee of $70 is charged, then a deposit of $85 arrives. A second fee of $35 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "188", "answer_3": "221", "answer_1": "207", "answer_0": "199"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $103. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "48", "answer_2": "16", "answer_0": "64", "answer_1": "22"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $40k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1959. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "67", "age_2": "66", "age_1": "64", "age_0": "65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "45", "answer_2": "40", "answer_0": "80", "answer_3": "40"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p1": "0.150", "p2": "0.050", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=902, unit_price=571, discount_pct=0, region=EU, sku_total=1473 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1473", "v3": "515944", "v2": "51504", "v0": "515042"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $226 per year. It increases by 15% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "192.1", "answer_3": "231.9", "answer_1": "259.9", "answer_0": "237.9"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $675. A fee of $54 is charged, then a deposit of $86 arrives. A second fee of $27 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "665", "answer_1": "682", "answer_2": "696", "answer_0": "680"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed January 26, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 29", "date_2": "January 31", "date_3": "January 31", "date_1": "January 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 811:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=621, unit_price=936, discount_pct=0, region=EU, sku_total=1557 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "581256", "v1": "1557", "v2": "58125", "v3": "581877"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "27", "answer_2": "24", "answer_3": "24", "answer_0": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed July 03, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 08", "date_0": "July 08", "date_2": "July 06", "date_1": "July 06"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 491:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $626. A fee of $126 is charged, then a deposit of $66 arrives. A second fee of $63 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "503", "answer_3": "511", "answer_1": "507", "answer_2": "493"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = $671; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed August 17, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 24", "date_0": "August 26", "date_3": "August 26", "date_2": "August 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 10 people. A shopper bought a 70 gram box and a 700 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $378. A fee of $60 is charged, then a deposit of $51 arrives. A second fee of $30 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "339", "answer_2": "328", "answer_1": "330", "answer_3": "355"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Wednesday, October 14, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = $383; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "42", "answer_2": "14", "answer_1": "18", "answer_3": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1986. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "41", "age_1": "39", "age_3": "38", "age_0": "40"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.333", "p3": "0.625", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = $169; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $530. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, October 01, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $281. A fee of $125 is charged, then a deposit of $24 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "136", "answer_2": "112", "answer_3": "120", "answer_0": "118"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.375", "p1": "0.500", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 525:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = 2026-06-04; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed September 11, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 18", "date_2": "September 16", "date_1": "September 16", "date_0": "September 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, July 17, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = $108; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.222", "p3": "0.125", "p1": "0.778", "p2": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 09:45 to 10:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h00", "dur_3": "0h00", "dur_0": "0h30", "dur_1": "1h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "20", "answer_3": "10", "answer_2": "10", "answer_1": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=860, unit_price=776, discount_pct=0, region=EU, sku_total=1636 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "668220", "v2": "66736", "v0": "667360", "v1": "1636"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed January 05, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 13", "date_2": "January 13", "date_3": "January 15", "date_0": "January 15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 11 people. A shopper bought a 44 gram box and a 440 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.800", "p0": "0.200", "p2": "0.800", "p3": "0.111"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 99 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "125", "answer_1": "97", "answer_0": "108", "answer_3": "100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed July 03, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 09", "date_0": "July 13", "date_3": "July 11", "date_2": "July 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 11:30 to 12:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "1h00", "dur_3": "0h15", "dur_1": "1h15", "dur_2": "1h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Friday, May 15, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.800", "p0": "0.200", "p3": "0.111", "p1": "0.800"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1965. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "62", "age_3": "59", "age_1": "60", "age_0": "61"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 82 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_1": "104", "answer_3": "134", "answer_2": "108"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, July 19, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $557. A fee of $192 is charged, then a deposit of $75 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "370", "answer_0": "344", "answer_1": "327", "answer_3": "319"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p2": "0.250", "p3": "0.083", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $396. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.462", "p1": "0.538", "p3": "0.417", "p2": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = 2026-05-28; priority = high.\n\nClaim to verify: \"the status is 2026-05-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p0": "0.500", "p1": "0.625", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1557 per year. It increases by 5% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1595.85", "answer_0": "1590.85", "answer_1": "1634.85", "answer_2": "1479.15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 749:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "30", "answer_2": "30", "answer_0": "60", "answer_1": "35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $604. A fee of $43 is charged, then a deposit of $78 arrives. A second fee of $21 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "609", "answer_0": "618", "answer_3": "602", "answer_1": "589"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p2": "0.100", "p3": "0.050", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $315. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1989. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "34", "age_0": "36", "age_2": "37", "age_1": "35"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, July 13, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1974. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "48", "age_1": "49", "age_2": "51", "age_0": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 325 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "295", "answer_1": "282", "answer_0": "288", "answer_3": "308"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 162 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "25", "answer_3": "58", "answer_0": "40", "answer_2": "20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 7 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1987. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "39", "age_3": "41", "age_1": "38", "age_2": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 19:15 to 22:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h45", "dur_3": "2h00", "dur_2": "3h00", "dur_1": "3h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed August 05, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "August 09", "date_3": "August 11", "date_0": "August 11", "date_1": "August 09"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = $222; priority = low.\n\nClaim to verify: \"the amount is $222\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "70", "answer_1": "34", "answer_3": "42", "answer_2": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "21", "answer_1": "25", "answer_2": "18", "answer_0": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 120 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "81", "answer_1": "78", "answer_0": "96", "answer_3": "106"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_2": "15", "answer_3": "21", "answer_1": "22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1593 per year. It increases by 15% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1785.95", "answer_2": "1354.05", "answer_3": "1761.95", "answer_1": "1831.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1617 per year. It increases by 15% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1859.55", "answer_0": "1841.55", "answer_3": "1878.55", "answer_2": "1374.45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 85 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "39", "answer_3": "64", "answer_0": "56", "answer_1": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1971. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "55", "age_0": "54", "age_1": "53", "age_3": "52"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = $397; priority = low.\n\nClaim to verify: \"the amount is $397\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Monday, September 14, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $25k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 4 grams of salt per serving and serves 5 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 693:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = 2026-08-09; priority = low.\n\nClaim to verify: \"the date is 2026-08-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, November 27, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 14:15 to 20:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "5h45", "dur_3": "5h30", "dur_1": "6h30", "dur_2": "6h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 11 people. A shopper bought a 66 gram box and a 660 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 791:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $426. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $107. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 63 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "229", "answer_1": "215", "answer_0": "224", "answer_2": "235"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 263 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "234", "answer_2": "243", "answer_3": "263", "answer_0": "252"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $642 per year. It increases by 5% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "631.1", "answer_1": "674.1", "answer_0": "658.1", "answer_2": "609.9"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = 2026-10-12; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $609. A fee of $174 is charged, then a deposit of $21 arrives. A second fee of $87 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "392", "answer_1": "363", "answer_3": "397", "answer_0": "369"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = open; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1986. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "40", "age_1": "37", "age_2": "39", "age_0": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 374 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "305", "answer_0": "288", "answer_1": "291", "answer_2": "279"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed November 13, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 24", "date_1": "November 22", "date_3": "November 24", "date_0": "November 26"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p0": "0.667", "p3": "0.167", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=935, unit_price=208, discount_pct=0, region=EU, sku_total=1143 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "19448", "v1": "1143", "v0": "194480", "v3": "195415"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, March 04, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $235 per year. It increases by 5% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "246.75", "answer_2": "223.25", "answer_3": "220.75", "answer_0": "230.75"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.615", "p3": "0.583", "p2": "0.385", "p1": "0.385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 283 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "75", "answer_3": "69", "answer_2": "69", "answer_0": "80"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = $609; priority = low.\n\nClaim to verify: \"the amount is $609\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p3": "0.625", "p0": "0.667", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=870, unit_price=788, discount_pct=0, region=EU, sku_total=1658 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "685560", "v3": "684690", "v1": "1658", "v2": "68556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = 2026-07-20; priority = low.\n\nClaim to verify: \"the owner is 2026-07-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1542 per year. It increases by 10% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1387.8", "answer_1": "1696.2", "answer_3": "1685.2", "answer_0": "1660.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = 2026-04-16; priority = high.\n\nClaim to verify: \"the owner is 2026-04-16\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1976. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "47", "age_0": "48", "age_2": "49", "age_3": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, May 26, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed September 07, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 11", "date_0": "September 11", "date_2": "September 09", "date_3": "September 13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, March 24, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=347, unit_price=114, discount_pct=0, region=EU, sku_total=461 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "461", "v0": "39558", "v3": "39211", "v2": "3955"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = 2026-10-28; priority = low.\n\nClaim to verify: \"the amount is 2026-10-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 09:00 to 10:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "1h00", "dur_3": "0h00", "dur_2": "0h45", "dur_0": "0h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.667", "p3": "0.273", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $16k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $564. A fee of $138 is charged, then a deposit of $90 arrives. A second fee of $69 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "420", "answer_3": "427", "answer_2": "426", "answer_0": "447"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.800", "p3": "0.111", "p1": "0.800", "p0": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $28k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $265. A fee of $76 is charged, then a deposit of $35 arrives. A second fee of $38 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "209", "answer_1": "170", "answer_0": "186", "answer_2": "209"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 18:00 to 20:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h00", "dur_3": "1h15", "dur_2": "2h30", "dur_1": "2h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $868 per year. It increases by 25% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1055", "answer_0": "1048", "answer_1": "1085", "answer_2": "651"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = 2026-12-09; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = 2026-07-06; priority = high.\n\nClaim to verify: \"the date is 2026-07-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_0": "30", "answer_1": "22", "answer_3": "12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $453 per year. It increases by 25% this year, then a $49 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "566.25", "answer_0": "517.25", "answer_3": "535.25", "answer_2": "339.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1973 per year. It increases by 5% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2061.65", "answer_2": "1874.35", "answer_1": "2071.65", "answer_3": "2022.65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = $697; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1116 per year. It increases by 25% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "837", "answer_0": "1359", "answer_3": "1342", "answer_1": "1395"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 249 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "79", "answer_1": "80", "answer_0": "60", "answer_2": "65"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=844, unit_price=736, discount_pct=0, region=EU, sku_total=1580 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "62118", "v0": "621184", "v1": "1580", "v3": "620340"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $555 per year. It increases by 5% this year, then a $49 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "499.75", "answer_2": "527.25", "answer_0": "533.75", "answer_1": "582.75"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = closed; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.150", "p0": "0.200", "p3": "0.050"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $288. A fee of $75 is charged, then a deposit of $36 arrives. A second fee of $37 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "236", "answer_1": "203", "answer_0": "212", "answer_3": "204"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Thursday, March 12, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Thursday, March 19, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=403, unit_price=268, discount_pct=0, region=EU, sku_total=671 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "108004", "v1": "671", "v3": "107601", "v2": "10800"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1313 per year. It increases by 5% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1247.35", "answer_3": "1381.65", "answer_1": "1378.65", "answer_0": "1354.65"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=716, unit_price=750, discount_pct=0, region=EU, sku_total=1466 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "537716", "v2": "53700", "v0": "537000", "v1": "1466"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = $196; priority = high.\n\nClaim to verify: \"the date is $196\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1982. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "46", "age_0": "44", "age_2": "45", "age_1": "43"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 07:30 to 11:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "4h15", "dur_3": "3h15", "dur_1": "4h15", "dur_0": "4h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 178 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "199", "answer_0": "192", "answer_3": "197", "answer_2": "181"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed October 18, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 23", "date_1": "October 21", "date_2": "October 23", "date_0": "October 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "70", "answer_3": "35", "answer_2": "35", "answer_1": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1960. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "67", "age_1": "65", "age_0": "66", "age_3": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=410, unit_price=561, discount_pct=0, region=EU, sku_total=971 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "23001", "v1": "971", "v0": "230010", "v3": "229600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1968. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_3": "60", "age_2": "59", "age_1": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $458. A fee of $75 is charged, then a deposit of $21 arrives. A second fee of $37 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "362", "answer_2": "387", "answer_0": "367", "answer_3": "340"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "17", "answer_3": "21", "answer_0": "35", "answer_2": "14"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = $548; priority = high.\n\nClaim to verify: \"the owner is $548\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1958. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "68", "age_1": "65", "age_2": "67", "age_0": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Saturday, August 08, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=710, unit_price=374, discount_pct=0, region=EU, sku_total=1084 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "264830", "v2": "26554", "v1": "1084", "v0": "265540"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p2": "0.571", "p0": "0.429", "p1": "0.571"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = $516; priority = high.\n\nClaim to verify: \"the status is $516\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = pending; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Thursday, January 01, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 10:45 to 17:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "6h45", "dur_3": "6h00", "dur_0": "6h15", "dur_1": "7h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $743. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $800. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1980. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "43", "age_3": "42", "age_0": "44", "age_2": "45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "20", "answer_2": "15", "answer_1": "19", "answer_0": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $843. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 16:15 to 17:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "0h45", "dur_3": "0h00", "dur_1": "1h00", "dur_2": "1h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=335, unit_price=276, discount_pct=0, region=EU, sku_total=611 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "611", "v3": "92795", "v2": "9246", "v0": "92460"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p3": "0.167", "p2": "0.500", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 385 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "122", "answer_0": "126", "answer_3": "117", "answer_2": "128"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = 2026-08-06; priority = low.\n\nClaim to verify: \"the status is 2026-08-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 911:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.222", "p0": "0.778", "p3": "0.750", "p2": "0.222"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_3": "20", "answer_2": "20", "answer_1": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 279 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_3": "50", "answer_1": "21", "answer_2": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = $533; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p0": "0.150", "p1": "0.200", "p3": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.375", "p0": "0.444", "p2": "0.556", "p1": "0.556"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A recipe needs 7 grams of salt per serving and serves 4 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "8", "answer_0": "16", "answer_1": "10", "answer_3": "8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1987. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "37", "age_2": "39", "age_3": "36", "age_0": "38"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = open; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, March 07, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 386:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.417", "p1": "0.583", "p3": "0.364", "p2": "0.583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.667", "p0": "0.500", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p0": "0.150", "p3": "0.050", "p1": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: status = ops; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = 2026-11-25; priority = high.\n\nClaim to verify: \"the owner is 2026-11-25\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.125", "p0": "0.250", "p1": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1986. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "39", "age_1": "37", "age_3": "40", "age_0": "38"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Monday, May 18, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed December 12, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 20", "date_0": "December 21", "date_1": "December 18", "date_2": "December 23"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=915, unit_price=561, discount_pct=0, region=EU, sku_total=1476 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1476", "v2": "51331", "v3": "514230", "v0": "513315"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = 2026-09-21; priority = low.\n\nClaim to verify: \"the status is 2026-09-21\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=315, unit_price=328, discount_pct=0, region=EU, sku_total=643 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "103635", "v0": "103320", "v1": "643", "v2": "10332"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1985. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "42", "age_1": "40", "age_0": "41", "age_3": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $578. A fee of $49 is charged, then a deposit of $65 arrives. A second fee of $24 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "570", "answer_1": "557", "answer_3": "580", "answer_2": "559"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 377 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "32", "answer_0": "48", "answer_3": "41", "answer_1": "31"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 170 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "33", "answer_2": "57", "answer_1": "59", "answer_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 5 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.556", "p1": "0.400", "p2": "0.400", "p0": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $286. A fee of $53 is charged, then a deposit of $53 arrives. A second fee of $26 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "260", "answer_1": "272", "answer_3": "266", "answer_2": "233"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $786. A fee of $168 is charged, then a deposit of $59 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "595", "answer_0": "593", "answer_2": "574", "answer_3": "583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1971. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "54", "age_0": "55", "age_2": "56", "age_3": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "25", "answer_2": "18", "answer_3": "42", "answer_0": "60"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = 2026-10-09; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = 2026-02-15; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1974 per year. It increases by 20% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2291.8", "answer_1": "2368.8", "answer_0": "2321.8", "answer_2": "1579.2"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 639:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed July 20, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 30", "date_3": "July 30", "date_1": "July 28", "date_2": "August 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.429", "p1": "0.500", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $498. A fee of $84 is charged, then a deposit of $23 arrives. A second fee of $42 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "395", "answer_1": "379", "answer_2": "408", "answer_3": "388"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $488. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed February 02, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 13", "date_3": "February 11", "date_0": "February 11", "date_1": "February 09"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, February 24, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p1": "0.800", "p2": "0.800", "p3": "0.111"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = 2026-08-24; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1988. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "38", "age_3": "35", "age_1": "36", "age_0": "37"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $124. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 88 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "81", "answer_2": "100", "answer_1": "86", "answer_0": "84"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.455", "p2": "0.500", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 07:45 to 10:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "1h45", "dur_0": "2h15", "dur_1": "2h45", "dur_2": "2h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, December 27, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "28", "answer_1": "31", "answer_2": "24", "answer_0": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1977. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "46", "age_3": "45", "age_0": "47", "age_2": "48"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1968. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "55", "age_0": "56", "age_3": "54", "age_2": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $614. A fee of $61 is charged, then a deposit of $70 arrives. A second fee of $30 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "569", "answer_2": "616", "answer_0": "593", "answer_3": "591"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 2 grams of salt per serving and serves 9 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=304, unit_price=655, discount_pct=0, region=EU, sku_total=959 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "959", "v2": "19912", "v0": "199120", "v3": "199424"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.083", "p1": "0.333", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1980. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "43", "age_0": "45", "age_1": "44", "age_2": "46"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = $510; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 256 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "229", "answer_3": "262", "answer_0": "245", "answer_1": "232"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 3 grams of salt per serving and serves 3 people. A shopper bought a 9 gram box and a 90 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_2": "30", "answer_3": "30", "answer_1": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 7 grams of salt per serving and serves 5 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 6 grams of salt per serving and serves 12 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 262 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "51", "answer_0": "60", "answer_2": "43", "answer_3": "79"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.364", "p1": "0.636", "p3": "0.300", "p2": "0.636"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.444", "p3": "0.375", "p1": "0.556", "p2": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1964. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "61", "age_2": "63", "age_3": "60", "age_0": "62"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $550 per year. It increases by 20% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "660", "answer_0": "614", "answer_3": "600", "answer_2": "440"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = $262; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.571", "p2": "0.375", "p0": "0.625", "p1": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=103, unit_price=457, discount_pct=0, region=EU, sku_total=560 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "46968", "v0": "47071", "v2": "4707", "v1": "560"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: date = $617; priority = low.\n\nClaim to verify: \"the date is $617\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = $875; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = ops; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 5 grams of salt per serving and serves 9 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $504. A fee of $145 is charged, then a deposit of $38 arrives. A second fee of $72 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "325", "answer_2": "323", "answer_3": "320", "answer_1": "339"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = 2026-09-27; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_0": "36", "answer_1": "16", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "20", "answer_0": "55", "answer_3": "35", "answer_1": "27"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=364, unit_price=905, discount_pct=0, region=EU, sku_total=1269 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "32942", "v3": "329056", "v0": "329420", "v1": "1269"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = pending; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $314 per year. It increases by 25% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "389.5", "answer_0": "363.5", "answer_1": "392.5", "answer_2": "235.5"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.500", "p0": "0.333", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Tuesday, January 20, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p3": "0.050", "p2": "0.050", "p0": "0.100"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, September 21, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.750", "p2": "0.750", "p0": "0.250", "p3": "0.143"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = $354; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 139 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "65", "answer_0": "48", "answer_2": "61", "answer_1": "57"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1977. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "49", "age_0": "47", "age_2": "48", "age_1": "46"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.200", "p1": "0.667", "p2": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $353 per year. It increases by 10% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "343.3", "answer_2": "317.7", "answer_1": "388.3", "answer_3": "375.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed May 09, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 17", "date_0": "May 19", "date_3": "May 18", "date_1": "May 16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed November 21, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 04", "date_1": "December 01", "date_2": "December 02", "date_3": "December 03"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = 2026-04-26; priority = low.\n\nClaim to verify: \"the amount is 2026-04-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "49", "answer_2": "35", "answer_1": "42", "answer_0": "84"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = $582; priority = low.\n\nClaim to verify: \"the status is $582\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed June 10, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "June 19", "date_2": "June 21", "date_1": "June 17", "date_3": "June 19"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 09:00 to 10:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "0h30", "dur_0": "1h15", "dur_2": "1h30", "dur_1": "1h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 742:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Sunday, August 02, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $845 per year. It increases by 25% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1056.25", "answer_0": "1011.25", "answer_3": "1003.25", "answer_2": "633.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = $648; priority = low.\n\nClaim to verify: \"the status is $648\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.750", "p2": "0.750", "p3": "0.143"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed July 21, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 31", "date_0": "August 04", "date_3": "August 02", "date_2": "August 06"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "21", "answer_3": "20", "answer_0": "36", "answer_2": "16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $753 per year. It increases by 25% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "914.25", "answer_1": "941.25", "answer_0": "924.25", "answer_2": "564.75"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1200 per year. It increases by 20% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "960", "answer_3": "1414", "answer_0": "1423", "answer_1": "1440"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p3": "0.050", "p1": "0.200", "p0": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed December 06, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 15", "date_1": "December 13", "date_2": "December 17", "date_0": "December 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $257. A fee of $191 is charged, then a deposit of $73 arrives. A second fee of $95 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "60", "answer_0": "44", "answer_2": "36", "answer_1": "17"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = 2026-11-23; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.125", "p3": "0.125", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.364", "p2": "0.636", "p1": "0.636", "p3": "0.300"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=671, unit_price=872, discount_pct=0, region=EU, sku_total=1543 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "585783", "v1": "1543", "v2": "58511", "v0": "585112"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.167", "p1": "0.333", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p3": "0.050", "p1": "0.150", "p2": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $522. A fee of $41 is charged, then a deposit of $32 arrives. A second fee of $20 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "513", "answer_2": "515", "answer_3": "499", "answer_0": "493"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = $350; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1500 per year. It increases by 10% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1350", "answer_1": "1650", "answer_3": "1672", "answer_0": "1637"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 369 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "122", "answer_1": "138", "answer_0": "126", "answer_2": "106"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 55 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "240", "answer_2": "231", "answer_3": "227", "answer_1": "253"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 6 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, August 15, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, April 05, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 275 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "90", "answer_2": "82", "answer_1": "94", "answer_3": "107"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 285:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 839:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 10:45 to 17:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "5h30", "dur_2": "6h45", "dur_0": "6h15", "dur_1": "6h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.273", "p2": "0.273", "p3": "0.700", "p0": "0.727"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 18:45 to 22:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "2h30", "dur_0": "3h15", "dur_2": "3h45", "dur_1": "3h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.444", "p2": "0.444", "p3": "0.500", "p0": "0.556"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = 2026-10-23; priority = high.\n\nClaim to verify: \"the date is 2026-10-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $948 per year. It increases by 10% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1006.8", "answer_2": "853.2", "answer_1": "1042.8", "answer_3": "968.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1971. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "56", "age_3": "53", "age_1": "54", "age_0": "55"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed October 26, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 06", "date_1": "November 02", "date_0": "November 04", "date_3": "November 04"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 167 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "137", "answer_1": "142", "answer_2": "131", "answer_0": "125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 823:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "48", "answer_1": "54", "answer_3": "48", "answer_0": "96"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.500", "p2": "0.500", "p3": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_2": "24", "answer_3": "24", "answer_1": "27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = 2026-02-15; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $2k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = ops; priority = low.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: owner = $163; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "35", "answer_2": "14", "answer_1": "17", "answer_3": "21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 5 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 79 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "240", "answer_3": "260", "answer_1": "245", "answer_2": "222"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $733 per year. It increases by 20% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "586.4", "answer_3": "869.6", "answer_0": "837.6", "answer_1": "879.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.538", "p2": "0.538", "p0": "0.462", "p3": "0.417"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 332 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "36", "answer_1": "73", "answer_0": "56", "answer_2": "40"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p3": "0.125", "p0": "0.250", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 777:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=254, unit_price=863, discount_pct=0, region=EU, sku_total=1117 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "218948", "v0": "219202", "v1": "1117", "v2": "21920"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 384 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "100", "answer_0": "90", "answer_3": "108", "answer_2": "81"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = 2026-05-21; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 68 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "133", "answer_0": "140", "answer_1": "147", "answer_3": "138"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.500", "p3": "0.167", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_1": "12", "answer_2": "8", "answer_3": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $525. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_1": "22", "answer_2": "16", "answer_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 481:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.538", "p0": "0.462", "p2": "0.538", "p3": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 416:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.667", "p0": "0.333", "p3": "0.273"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.167", "p1": "0.250", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed March 20, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 26", "date_3": "March 26", "date_2": "March 28", "date_1": "March 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1974. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "53", "age_0": "51", "age_1": "50", "age_2": "52"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 4 people. A shopper bought a 8 gram box and a 80 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 413:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_2": "24", "answer_3": "24", "answer_1": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.500", "p3": "0.167", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 262 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "67", "answer_2": "67", "answer_0": "84", "answer_3": "86"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $238. A fee of $88 is charged, then a deposit of $70 arrives. A second fee of $44 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "148", "answer_1": "147", "answer_0": "176", "answer_2": "147"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.375", "p0": "0.250", "p2": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, January 23, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = 2026-01-20; priority = low.\n\nClaim to verify: \"the date is 2026-01-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 603:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, August 14, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 268 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "208", "answer_0": "225", "answer_2": "219", "answer_3": "242"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_1": "28", "answer_3": "15", "answer_2": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1988. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "35", "age_0": "36", "age_2": "37", "age_3": "34"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=197, unit_price=622, discount_pct=0, region=EU, sku_total=819 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "122534", "v1": "819", "v2": "12253", "v3": "122337"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = $690; priority = low.\n\nClaim to verify: \"the owner is $690\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1983. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "41", "age_2": "42", "age_1": "40", "age_3": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $638. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p3": "0.462", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 389:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = 2026-07-02; priority = low.\n\nClaim to verify: \"the amount is 2026-07-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1967. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "56", "age_1": "57", "age_0": "58", "age_2": "59"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $872 per year. It increases by 20% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1046.4", "answer_2": "697.6", "answer_3": "1026.4", "answer_0": "1017.4"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.750", "p3": "0.143", "p1": "0.750", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = 2026-06-03; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed February 06, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 19", "date_0": "February 17", "date_1": "February 13", "date_3": "February 15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed April 17, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 30", "date_2": "April 28", "date_1": "April 26", "date_3": "April 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_3": "30", "answer_0": "60", "answer_1": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = $762; priority = low.\n\nClaim to verify: \"the owner is $762\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $745 per year. It increases by 10% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "805.5", "answer_3": "766.5", "answer_2": "670.5", "answer_1": "819.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "43", "answer_0": "64", "answer_3": "24", "answer_2": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $508. A fee of $140 is charged, then a deposit of $84 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "403", "answer_0": "382", "answer_3": "352", "answer_2": "409"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.750", "p3": "0.143", "p0": "0.250", "p1": "0.750"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=150, unit_price=163, discount_pct=0, region=EU, sku_total=313 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "313", "v0": "24450", "v2": "2445", "v3": "24300"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.333", "p2": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "14", "answer_1": "21", "answer_0": "63", "answer_3": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, May 04, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = $922; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Saturday, September 19, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = 2026-11-02; priority = low.\n\nClaim to verify: \"the owner is 2026-11-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $525. A fee of $65 is charged, then a deposit of $23 arrives. A second fee of $32 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "421", "answer_0": "451", "answer_3": "473", "answer_1": "447"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed December 19, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 24", "date_3": "December 26", "date_2": "December 23", "date_0": "December 25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $205. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "18", "answer_2": "15", "answer_0": "30", "answer_3": "15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $679 per year. It increases by 10% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "611.1", "answer_1": "746.9", "answer_3": "698.9", "answer_0": "728.9"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1990. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "37", "age_1": "35", "age_3": "38", "age_0": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.364", "p3": "0.300", "p1": "0.636", "p2": "0.636"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = $256; priority = high.\n\nClaim to verify: \"the date is $256\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p3": "0.462", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "15", "answer_0": "25", "answer_1": "17", "answer_3": "10"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $585. A fee of $182 is charged, then a deposit of $86 arrives. A second fee of $91 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "382", "answer_0": "398", "answer_3": "410", "answer_1": "426"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed October 17, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 24", "date_3": "October 23", "date_0": "October 22", "date_1": "October 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.444", "p2": "0.444", "p0": "0.556", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=182, unit_price=627, discount_pct=0, region=EU, sku_total=809 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "809", "v3": "114296", "v2": "11411", "v0": "114114"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "35", "answer_1": "15", "answer_3": "25", "answer_2": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 08:15 to 19:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "10h45", "dur_3": "10h00", "dur_2": "11h15", "dur_1": "11h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "64", "answer_2": "32", "answer_3": "32", "answer_1": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.625", "p2": "0.333", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 5 grams of salt per serving and serves 3 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = $373; priority = high.\n\nClaim to verify: \"the amount is $373\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=605, unit_price=874, discount_pct=0, region=EU, sku_total=1479 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "528770", "v3": "528165", "v1": "1479", "v2": "52877"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1976. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "49", "age_3": "51", "age_1": "48", "age_2": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Friday, April 10, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $22k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = 2026-11-03; priority = high.\n\nClaim to verify: \"the status is 2026-11-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 295:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 08:30 to 20:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "11h15", "dur_0": "12h00", "dur_1": "12h15", "dur_2": "12h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $313. A fee of $183 is charged, then a deposit of $29 arrives. A second fee of $91 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "68", "answer_2": "94", "answer_1": "77", "answer_3": "92"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 314:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $14k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p0": "0.500", "p3": "0.125", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 240:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 139 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "168", "answer_1": "156", "answer_2": "158", "answer_3": "164"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1712 per year. It increases by 5% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1797.6", "answer_0": "1784.6", "answer_2": "1626.4", "answer_3": "1760.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "40", "answer_0": "80", "answer_1": "45", "answer_2": "40"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "77", "answer_1": "47", "answer_2": "42", "answer_3": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "28", "answer_1": "46", "answer_0": "70", "answer_2": "42"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1959. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "63", "age_0": "65", "age_1": "64", "age_2": "66"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 204 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "32", "answer_0": "36", "answer_3": "16", "answer_1": "38"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, February 08, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed April 10, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 13", "date_0": "April 15", "date_2": "April 17", "date_3": "April 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed July 06, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 15", "date_3": "July 15", "date_2": "July 17", "date_1": "July 13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.538", "p3": "0.417", "p1": "0.538", "p0": "0.462"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.625", "p1": "0.625", "p3": "0.286", "p0": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $292 per year. It increases by 25% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "318", "answer_2": "219", "answer_1": "365", "answer_3": "356"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 06:30 to 19:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "13h00", "dur_2": "13h00", "dur_0": "12h30", "dur_3": "12h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 341 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "270", "answer_3": "268", "answer_2": "280", "answer_1": "284"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "10", "answer_3": "30", "answer_1": "16", "answer_0": "40"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $27k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $664. A fee of $192 is charged, then a deposit of $85 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "461", "answer_2": "456", "answer_1": "489", "answer_3": "484"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 252 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "40", "answer_1": "43", "answer_0": "36", "answer_2": "42"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 3 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 690:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=584, unit_price=655, discount_pct=0, region=EU, sku_total=1239 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "382520", "v2": "38252", "v3": "383104", "v1": "1239"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = closed; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed February 14, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 18", "date_0": "February 19", "date_2": "February 21", "date_3": "February 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1305 per year. It increases by 10% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1435.5", "answer_3": "1436.5", "answer_0": "1406.5", "answer_2": "1174.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1971. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "55", "age_3": "52", "age_1": "53", "age_0": "54"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $674 per year. It increases by 25% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "822.5", "answer_0": "813.5", "answer_2": "505.5", "answer_1": "842.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = $619; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.533", "p2": "0.533", "p3": "0.429", "p0": "0.467"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed October 05, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 11", "date_1": "October 09", "date_3": "October 11", "date_0": "October 09"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 80 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_1": "58", "answer_3": "91", "answer_2": "85"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.167", "p0": "0.333", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.375", "p2": "0.556", "p1": "0.556", "p0": "0.444"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $30k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 653:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = 2026-10-18; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: status = $469; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.250", "p1": "0.500", "p0": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $25k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $663. A fee of $195 is charged, then a deposit of $64 arrives. A second fee of $97 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "440", "answer_0": "435", "answer_2": "440", "answer_1": "426"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 964:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=804, unit_price=885, discount_pct=0, region=EU, sku_total=1689 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "710736", "v1": "1689", "v0": "711540", "v2": "71154"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $337. A fee of $158 is charged, then a deposit of $55 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "128", "answer_2": "142", "answer_0": "155", "answer_3": "181"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=532, unit_price=869, discount_pct=0, region=EU, sku_total=1401 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "46230", "v0": "462308", "v1": "1401", "v3": "462840"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.667", "p1": "0.833", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = $53; priority = high.\n\nClaim to verify: \"the status is $53\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = closed; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 501:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = 2026-11-23; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 236 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "76", "answer_3": "65", "answer_0": "63", "answer_2": "77"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 518:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=625, unit_price=979, discount_pct=0, region=EU, sku_total=1604 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "61187", "v3": "612500", "v0": "611875", "v1": "1604"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "36", "answer_2": "12", "answer_0": "48", "answer_1": "18"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=878, unit_price=755, discount_pct=0, region=EU, sku_total=1633 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "663768", "v1": "1633", "v2": "66289", "v0": "662890"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $565. A fee of $154 is charged, then a deposit of $81 arrives. A second fee of $77 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "415", "answer_3": "401", "answer_1": "402", "answer_2": "386"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $21k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1390 per year. It increases by 25% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1711.5", "answer_1": "1737.5", "answer_2": "1042.5", "answer_3": "1722.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "17", "answer_0": "25", "answer_3": "10", "answer_2": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_1": "21", "answer_2": "18", "answer_3": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.250", "p2": "0.150", "p0": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed October 05, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "October 08", "date_3": "October 10", "date_1": "October 08", "date_2": "October 10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1677 per year. It increases by 20% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1926.4", "answer_2": "1341.6", "answer_0": "1966.4", "answer_1": "2012.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 166 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "270", "answer_3": "283", "answer_2": "259", "answer_1": "262"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 8 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 220 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "89", "answer_1": "97", "answer_2": "71", "answer_0": "80"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 261 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "131", "answer_0": "125", "answer_2": "117", "answer_1": "115"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, August 03, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $508. A fee of $75 is charged, then a deposit of $38 arrives. A second fee of $37 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "404", "answer_1": "456", "answer_3": "423", "answer_0": "434"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1606 per year. It increases by 10% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1718.6", "answer_3": "1732.6", "answer_1": "1766.6", "answer_2": "1445.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $355. A fee of $54 is charged, then a deposit of $38 arrives. A second fee of $27 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "312", "answer_2": "325", "answer_3": "287", "answer_1": "290"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $486 per year. It increases by 25% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "364.5", "answer_1": "607.5", "answer_0": "563.5", "answer_3": "524.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed July 08, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 19", "date_1": "July 17", "date_0": "July 21", "date_2": "July 23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed December 20, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 31", "date_3": "December 29", "date_0": "December 29", "date_1": "December 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=702, unit_price=243, discount_pct=0, region=EU, sku_total=945 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "17058", "v0": "170586", "v1": "945", "v3": "171288"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = 2026-05-10; priority = high.\n\nClaim to verify: \"the status is 2026-05-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 09:30 to 12:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h00", "dur_3": "2h00", "dur_0": "2h30", "dur_1": "3h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.167", "p3": "0.167", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 8 grams of salt per serving and serves 3 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 5 grams of salt per serving and serves 10 people. A shopper bought a 50 gram box and a 500 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.125", "p0": "0.375", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=222, unit_price=559, discount_pct=0, region=EU, sku_total=781 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "781", "v0": "124098", "v2": "12409", "v3": "123876"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = open; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = pending; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 388 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "160", "answer_2": "150", "answer_1": "126", "answer_0": "144"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.167", "p2": "0.083", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $366. A fee of $134 is charged, then a deposit of $67 arrives. A second fee of $67 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "253", "answer_2": "205", "answer_0": "232", "answer_3": "259"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "15", "answer_0": "21", "answer_2": "6", "answer_1": "11"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1987. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "38", "age_3": "39", "age_0": "37", "age_1": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = $243; priority = low.\n\nClaim to verify: \"the owner is $243\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 7 grams of salt per serving and serves 12 people. A shopper bought a 84 gram box and a 840 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=421, unit_price=153, discount_pct=0, region=EU, sku_total=574 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "64413", "v2": "6441", "v3": "64834", "v1": "574"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p2": "0.125", "p3": "0.125", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 11:00 to 15:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "4h30", "dur_2": "4h15", "dur_0": "4h00", "dur_3": "3h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 939:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = 2026-08-26; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p1": "0.200", "p3": "0.050", "p0": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = $824; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = $769; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = $653; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = 2026-02-09; priority = low.\n\nClaim to verify: \"the amount is 2026-02-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: status = 2026-06-01; priority = high.\n\nClaim to verify: \"the status is 2026-06-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 277:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed April 09, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 15", "date_3": "April 17", "date_2": "April 19", "date_0": "April 17"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $418. A fee of $111 is charged, then a deposit of $51 arrives. A second fee of $55 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "303", "answer_3": "314", "answer_2": "326", "answer_1": "277"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed December 23, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 06", "date_2": "January 08", "date_3": "January 04", "date_1": "January 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed April 02, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 12", "date_3": "April 10", "date_0": "April 10", "date_1": "April 08"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1979. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "47", "age_0": "46", "age_1": "45", "age_3": "44"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $867. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1960. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "65", "age_3": "64", "age_0": "66", "age_2": "67"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $15k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 377 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "111", "answer_0": "105", "answer_2": "87", "answer_1": "90"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $470. A fee of $96 is charged, then a deposit of $41 arrives. A second fee of $48 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "356", "answer_1": "378", "answer_2": "349", "answer_0": "367"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $420. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.125", "p1": "0.375", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 170 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "180", "answer_1": "157", "answer_0": "168", "answer_2": "153"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.800", "p3": "0.111", "p0": "0.200", "p2": "0.800"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=972, unit_price=844, discount_pct=0, region=EU, sku_total=1816 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1816", "v0": "820368", "v3": "821340", "v2": "82036"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "42", "answer_1": "17", "answer_2": "12", "answer_3": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $472 per year. It increases by 25% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "553", "answer_3": "536", "answer_1": "590", "answer_2": "354"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "9", "answer_0": "30", "answer_1": "16", "answer_3": "21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed March 17, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 29", "date_0": "March 27", "date_3": "March 27", "date_1": "March 25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, April 10, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.400", "p2": "0.545", "p1": "0.545", "p0": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $264. A fee of $159 is charged, then a deposit of $51 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "69", "answer_2": "83", "answer_3": "94", "answer_0": "77"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1965. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "60", "age_3": "63", "age_2": "62", "age_0": "61"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.667", "p3": "0.625", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = $375; priority = high.\n\nClaim to verify: \"the owner is $375\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 221 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "91", "answer_3": "56", "answer_1": "64", "answer_0": "72"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $527 per year. It increases by 10% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "474.3", "answer_0": "561.7", "answer_1": "579.7", "answer_3": "567.7"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 3 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Sunday, July 19, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1982. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "45", "age_2": "44", "age_0": "43", "age_1": "42"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = $213; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $645. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1980. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "45", "age_3": "42", "age_0": "44", "age_1": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 374:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, January 08, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.375", "p3": "0.125", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=277, unit_price=697, discount_pct=0, region=EU, sku_total=974 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "19306", "v1": "974", "v0": "193069", "v3": "193346"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 3 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 262 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "77", "answer_0": "80", "answer_2": "74", "answer_3": "77"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1928 per year. It increases by 5% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2030.4", "answer_0": "1995.4", "answer_1": "2024.4", "answer_2": "1831.6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1523 per year. It increases by 15% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1732.45", "answer_1": "1751.45", "answer_3": "1762.45", "answer_2": "1294.55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = $39; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "21", "answer_1": "27", "answer_0": "63", "answer_3": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, November 06, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = $944; priority = low.\n\nClaim to verify: \"the status is $944\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Sunday, December 27, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 14:30 to 17:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h30", "dur_0": "2h15", "dur_2": "2h30", "dur_3": "1h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $762. A fee of $95 is charged, then a deposit of $52 arrives. A second fee of $47 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "662", "answer_3": "645", "answer_0": "672", "answer_1": "664"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p3": "0.125", "p0": "0.250", "p2": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An account starts at $732. A fee of $177 is charged, then a deposit of $58 arrives. A second fee of $88 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "525", "answer_2": "546", "answer_3": "516", "answer_1": "534"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 10:30 to 11:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "0h00", "dur_1": "1h00", "dur_2": "1h00", "dur_0": "0h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed June 08, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 16", "date_3": "June 18", "date_0": "June 18", "date_1": "June 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p0": "0.333", "p1": "0.667", "p2": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed November 17, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 22", "date_3": "November 24", "date_0": "November 24", "date_2": "November 22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=422, unit_price=187, discount_pct=0, region=EU, sku_total=609 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "78492", "v0": "78914", "v2": "7891", "v1": "609"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=452, unit_price=683, discount_pct=0, region=EU, sku_total=1135 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "30871", "v1": "1135", "v3": "308264", "v0": "308716"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $25k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 311:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "20", "answer_2": "10", "answer_3": "10", "answer_1": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, July 12, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 10 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 313 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "78", "answer_2": "44", "answer_3": "70", "answer_0": "64"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p3": "0.167", "p2": "0.500", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $511. A fee of $62 is charged, then a deposit of $70 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "516", "answer_2": "503", "answer_0": "488", "answer_1": "469"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=899, unit_price=224, discount_pct=0, region=EU, sku_total=1123 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1123", "v3": "200477", "v2": "20137", "v0": "201376"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $803. A fee of $151 is charged, then a deposit of $73 arrives. A second fee of $75 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "650", "answer_1": "630", "answer_2": "621", "answer_3": "627"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $38k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.250", "p2": "0.083", "p0": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $442. A fee of $90 is charged, then a deposit of $34 arrives. A second fee of $45 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "316", "answer_0": "341", "answer_3": "345", "answer_1": "314"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 613:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 223 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "144", "answer_1": "156", "answer_2": "139", "answer_3": "142"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $791. A fee of $134 is charged, then a deposit of $44 arrives. A second fee of $67 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "634", "answer_3": "622", "answer_1": "664", "answer_2": "644"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 97 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "118", "answer_1": "112", "answer_3": "133", "answer_0": "120"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Friday, November 06, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $970 per year. It increases by 20% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "776", "answer_0": "1132", "answer_1": "1164", "answer_3": "1170"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, January 17, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "24", "answer_3": "18", "answer_0": "36", "answer_2": "18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.200", "p2": "0.150", "p1": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "10", "answer_0": "16", "answer_2": "8", "answer_3": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1982. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "42", "age_0": "43", "age_2": "44", "age_3": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = 2026-07-09; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $378. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = closed; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "24", "answer_3": "20", "answer_0": "40", "answer_2": "20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $354. A fee of $146 is charged, then a deposit of $85 arrives. A second fee of $73 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "220", "answer_2": "229", "answer_1": "206", "answer_3": "222"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Thursday, July 16, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1976. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "48", "age_2": "51", "age_0": "50", "age_1": "49"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 242:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p1": "0.250", "p2": "0.083", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $234. A fee of $191 is charged, then a deposit of $32 arrives. A second fee of $95 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "-13", "answer_3": "-40", "answer_1": "-14", "answer_0": "-20"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=738, unit_price=255, discount_pct=0, region=EU, sku_total=993 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "188190", "v3": "188928", "v1": "993", "v2": "18819"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 169 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "79", "answer_3": "79", "answer_0": "96", "answer_1": "94"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.385", "p3": "0.583", "p1": "0.385", "p0": "0.615"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, August 14, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 226:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed April 27, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 04", "date_2": "May 02", "date_1": "May 02", "date_3": "May 04"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.222", "p0": "0.300", "p1": "0.700", "p2": "0.700"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = $760; priority = high.\n\nClaim to verify: \"the date is $760\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = $546; priority = low.\n\nClaim to verify: \"the owner is $546\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.250", "p1": "0.500", "p0": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = 2026-10-07; priority = high.\n\nClaim to verify: \"the owner is 2026-10-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "32", "answer_0": "56", "answer_2": "24", "answer_1": "28"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = $937; priority = high.\n\nClaim to verify: \"the status is $937\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 9 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = 2026-09-20; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.600", "p3": "0.250", "p0": "0.400", "p1": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 382 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "168", "answer_0": "160", "answer_3": "180", "answer_1": "141"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = 2026-12-07; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = $462; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = $495; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.455", "p3": "0.500", "p0": "0.545", "p1": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 172 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "210", "answer_3": "206", "answer_1": "203", "answer_2": "228"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=941, unit_price=469, discount_pct=0, region=EU, sku_total=1410 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1410", "v2": "44132", "v0": "441329", "v3": "442270"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p3": "0.462", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 738:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 9 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = $671; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 4 grams of salt per serving and serves 3 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 07:15 to 13:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "5h30", "dur_2": "6h00", "dur_1": "6h00", "dur_3": "5h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed August 15, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 26", "date_2": "August 29", "date_1": "August 24", "date_0": "August 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p3": "0.462", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = 2026-01-22; priority = low.\n\nClaim to verify: \"the date is 2026-01-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=791, unit_price=485, discount_pct=0, region=EU, sku_total=1276 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "382844", "v1": "1276", "v0": "383635", "v2": "38363"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $588 per year. It increases by 10% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "653.8", "answer_2": "529.2", "answer_0": "621.8", "answer_1": "646.8"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 245:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = $886; priority = high.\n\nClaim to verify: \"the amount is $886\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 203 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "272", "answer_3": "273", "answer_0": "280", "answer_2": "292"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p1": "0.150", "p2": "0.050", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = $39; priority = low.\n\nClaim to verify: \"the amount is $39\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=461, unit_price=300, discount_pct=0, region=EU, sku_total=761 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "761", "v3": "137839", "v2": "13830", "v0": "138300"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=602, unit_price=510, discount_pct=0, region=EU, sku_total=1112 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "307622", "v2": "30702", "v1": "1112", "v0": "307020"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = 2026-07-05; priority = high.\n\nClaim to verify: \"the amount is 2026-07-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $859 per year. It increases by 5% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "906.95", "answer_1": "901.95", "answer_0": "867.95", "answer_2": "816.05"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = 2026-01-07; priority = high.\n\nClaim to verify: \"the date is 2026-01-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 263 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "140", "answer_1": "142", "answer_3": "151", "answer_0": "144"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = 2026-03-11; priority = low.\n\nClaim to verify: \"the owner is 2026-03-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1964. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "61", "age_2": "63", "age_0": "62", "age_3": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Saturday, July 11, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_2": "12", "answer_1": "15", "answer_0": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.444", "p2": "0.444", "p0": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_2": "16", "answer_0": "40", "answer_1": "19"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 70 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "68", "answer_0": "72", "answer_1": "86", "answer_3": "69"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.667", "p1": "0.300", "p2": "0.300", "p0": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1114 per year. It increases by 5% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1140.7", "answer_3": "1133.7", "answer_1": "1169.7", "answer_2": "1058.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $457. A fee of $199 is charged, then a deposit of $31 arrives. A second fee of $99 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "190", "answer_1": "175", "answer_2": "203", "answer_3": "181"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=808, unit_price=741, discount_pct=0, region=EU, sku_total=1549 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "598728", "v3": "597920", "v1": "1549", "v2": "59872"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p0": "0.500", "p3": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $308 per year. It increases by 20% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "338.6", "answer_3": "370.6", "answer_1": "369.6", "answer_2": "246.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Wednesday, April 22, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $757. A fee of $43 is charged, then a deposit of $38 arrives. A second fee of $21 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "722", "answer_2": "738", "answer_3": "701", "answer_0": "731"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=984, unit_price=435, discount_pct=0, region=EU, sku_total=1419 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "42804", "v3": "429024", "v1": "1419", "v0": "428040"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = 2026-06-17; priority = high.\n\nClaim to verify: \"the owner is 2026-06-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = pending; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.333", "p2": "0.167", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $880 per year. It increases by 25% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1082", "answer_2": "660", "answer_1": "1100", "answer_3": "1060"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.500", "p2": "0.375", "p1": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "18", "answer_1": "10", "answer_3": "12", "answer_2": "6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = 2026-04-25; priority = low.\n\nClaim to verify: \"the status is 2026-04-25\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.286", "p0": "0.714", "p2": "0.286", "p3": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = pending; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "22", "answer_2": "18", "answer_0": "30", "answer_3": "12"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.300", "p2": "0.300", "p0": "0.700", "p3": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $491. A fee of $55 is charged, then a deposit of $66 arrives. A second fee of $27 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "487", "answer_2": "483", "answer_0": "475", "answer_3": "458"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=537, unit_price=290, discount_pct=0, region=EU, sku_total=827 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "827", "v3": "156267", "v0": "155730", "v2": "15573"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 355 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "45", "answer_0": "30", "answer_3": "19", "answer_2": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p3": "0.050", "p2": "0.100", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "45", "answer_1": "17", "answer_2": "10", "answer_3": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=721, unit_price=619, discount_pct=0, region=EU, sku_total=1340 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "446299", "v2": "44629", "v1": "1340", "v3": "447020"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.417", "p0": "0.583", "p3": "0.545", "p1": "0.417"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1965. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "61", "age_1": "60", "age_2": "62", "age_3": "63"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1983. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "41", "age_3": "39", "age_2": "42", "age_1": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $869 per year. It increases by 5% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "907.45", "answer_1": "912.45", "answer_2": "825.55", "answer_0": "902.45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $889. A fee of $195 is charged, then a deposit of $28 arrives. A second fee of $97 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "625", "answer_2": "651", "answer_3": "622", "answer_1": "620"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.833", "p2": "0.500", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = $513; priority = low.\n\nClaim to verify: \"the date is $513\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_1": "52", "answer_3": "32", "answer_2": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = $504; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $627 per year. It increases by 5% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "595.65", "answer_0": "617.35", "answer_3": "590.35", "answer_1": "658.35"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.167", "p2": "0.333", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "15", "answer_0": "30", "answer_2": "15", "answer_1": "18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1958. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "68", "age_3": "66", "age_2": "69", "age_1": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $220. A fee of $106 is charged, then a deposit of $34 arrives. A second fee of $53 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "95", "answer_3": "108", "answer_2": "122", "answer_1": "78"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=443, unit_price=664, discount_pct=0, region=EU, sku_total=1107 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1107", "v3": "293709", "v0": "294152", "v2": "29415"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "42", "answer_1": "47", "answer_3": "35", "answer_0": "77"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed December 06, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 14", "date_0": "December 16", "date_1": "December 14", "date_3": "December 16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 297 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "78", "answer_3": "54", "answer_2": "48", "answer_0": "64"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = 2026-03-10; priority = low.\n\nClaim to verify: \"the status is 2026-03-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = $601; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1983. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "41", "age_2": "42", "age_1": "40", "age_3": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $781. A fee of $181 is charged, then a deposit of $27 arrives. A second fee of $90 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "523", "answer_3": "562", "answer_1": "524", "answer_0": "537"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Thursday, October 08, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = 2026-12-06; priority = low.\n\nClaim to verify: \"the date is 2026-12-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 192 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "108", "answer_2": "92", "answer_3": "92", "answer_1": "126"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = 2026-05-26; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, August 01, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.400", "p1": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, July 02, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 56 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_2": "40", "answer_3": "29", "answer_1": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p3": "0.636", "p0": "0.667", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Wednesday, September 02, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 4 grams of salt per serving and serves 3 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 928:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p0": "0.333", "p3": "0.200", "p1": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed September 10, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 19", "date_0": "September 21", "date_3": "September 19", "date_1": "September 17"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 5 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $285 per year. It increases by 25% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "299.25", "answer_1": "356.25", "answer_0": "331.25", "answer_2": "213.75"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, August 11, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.083", "p1": "0.417", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 730:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = $636; priority = high.\n\nClaim to verify: \"the date is $636\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $24k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 861:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1978. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "47", "age_2": "48", "age_3": "49", "age_1": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.833", "p3": "0.167", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "10", "answer_1": "15", "answer_3": "25", "answer_0": "35"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.083", "p2": "0.250", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = $652; priority = high.\n\nClaim to verify: \"the owner is $652\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed June 13, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "June 18", "date_2": "June 20", "date_3": "June 19", "date_1": "June 17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 3 grams of salt per serving and serves 10 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $624. A fee of $167 is charged, then a deposit of $76 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "480", "answer_2": "435", "answer_1": "468", "answer_0": "450"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $668. A fee of $33 is charged, then a deposit of $50 arrives. A second fee of $16 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "692", "answer_1": "664", "answer_3": "674", "answer_0": "669"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed March 21, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 31", "date_0": "April 03", "date_2": "April 05", "date_3": "April 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $599 per year. It increases by 25% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "697.75", "answer_1": "748.75", "answer_0": "705.75", "answer_2": "449.25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.750", "p0": "0.250", "p2": "0.750", "p3": "0.143"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $951 per year. It increases by 25% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "713.25", "answer_0": "1167.75", "answer_3": "1146.75", "answer_1": "1188.75"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $845. A fee of $99 is charged, then a deposit of $43 arrives. A second fee of $49 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "767", "answer_1": "727", "answer_0": "740", "answer_3": "710"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 2 grams of salt per serving and serves 9 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=787, unit_price=840, discount_pct=0, region=EU, sku_total=1627 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "661867", "v0": "661080", "v2": "66108", "v1": "1627"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Contract signed April 14, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 19", "date_3": "April 21", "date_0": "April 21", "date_1": "April 19"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "8", "answer_2": "12", "answer_0": "20", "answer_1": "14"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = 2026-10-27; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = 2026-10-01; priority = high.\n\nClaim to verify: \"the owner is 2026-10-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 349 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "65", "answer_3": "70", "answer_0": "84", "answer_2": "103"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p1": "0.150", "p0": "0.100", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 159 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "180", "answer_2": "197", "answer_1": "183", "answer_3": "183"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=473, unit_price=968, discount_pct=0, region=EU, sku_total=1441 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "457864", "v2": "45786", "v1": "1441", "v3": "458337"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.083", "p2": "0.083", "p0": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Saturday, November 07, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=351, unit_price=206, discount_pct=0, region=EU, sku_total=557 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "7230", "v0": "72306", "v1": "557", "v3": "71955"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.200", "p2": "0.100", "p0": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 429:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $32k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "38", "answer_3": "21", "answer_2": "35", "answer_0": "56"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1961. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "64", "age_0": "65", "age_3": "63", "age_2": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=921, unit_price=957, discount_pct=0, region=EU, sku_total=1878 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1878", "v3": "880476", "v0": "881397", "v2": "88139"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.150", "p3": "0.050", "p0": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $19k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1975. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "50", "age_0": "49", "age_1": "48", "age_3": "51"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, February 08, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = $469; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = $876; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 201 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "110", "answer_3": "107", "answer_2": "77", "answer_0": "90"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.375", "p1": "0.500", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1397 per year. It increases by 25% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1716.25", "answer_2": "1047.75", "answer_0": "1730.25", "answer_1": "1746.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=265, unit_price=159, discount_pct=0, region=EU, sku_total=424 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "424", "v3": "41870", "v2": "4213", "v0": "42135"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $22k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1970. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "57", "age_0": "56", "age_1": "55", "age_3": "58"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 306 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "87", "answer_2": "94", "answer_0": "100", "answer_1": "94"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = $735; priority = high.\n\nClaim to verify: \"the status is $735\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $434. A fee of $72 is charged, then a deposit of $57 arrives. A second fee of $36 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "407", "answer_0": "383", "answer_2": "386", "answer_1": "368"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 86 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "52", "answer_0": "36", "answer_2": "24", "answer_3": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $24k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = 2026-12-01; priority = high.\n\nClaim to verify: \"the date is 2026-12-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=602, unit_price=559, discount_pct=0, region=EU, sku_total=1161 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1161", "v0": "336518", "v3": "335916", "v2": "33651"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed September 06, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 14", "date_1": "September 12", "date_0": "September 14", "date_2": "September 12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $683. A fee of $107 is charged, then a deposit of $31 arrives. A second fee of $53 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "584", "answer_1": "549", "answer_0": "554", "answer_3": "552"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 718:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.222", "p0": "0.778", "p1": "0.222", "p3": "0.750"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 8 grams of salt per serving and serves 6 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Thursday, January 08, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1211 per year. It increases by 15% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1391.65", "answer_1": "1392.65", "answer_0": "1353.65", "answer_2": "1029.35"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.625", "p2": "0.333", "p0": "0.667", "p1": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = 2026-10-08; priority = high.\n\nClaim to verify: \"the owner is 2026-10-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $149. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1958. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "66", "age_2": "69", "age_0": "68", "age_1": "67"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.300", "p3": "0.667", "p2": "0.300", "p0": "0.700"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Wednesday, April 01, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = 2026-10-03; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $822. A fee of $167 is charged, then a deposit of $51 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "626", "answer_2": "600", "answer_1": "618", "answer_0": "623"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, November 16, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = $408; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1977. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "47", "age_0": "48", "age_2": "49", "age_3": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.667", "p1": "0.833", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1979. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "45", "age_1": "44", "age_2": "46", "age_3": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 222 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "21", "answer_1": "25", "answer_0": "36", "answer_2": "54"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_0": "42", "answer_3": "12", "answer_1": "32"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $230 per year. It increases by 20% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "184", "answer_0": "241", "answer_1": "276", "answer_3": "232"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=428, unit_price=227, discount_pct=0, region=EU, sku_total=655 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "9715", "v3": "97584", "v0": "97156", "v1": "655"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed January 06, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 16", "date_1": "January 14", "date_2": "January 18", "date_0": "January 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=161, unit_price=788, discount_pct=0, region=EU, sku_total=949 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "126868", "v2": "12686", "v3": "127029", "v1": "949"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A shift runs from 12:15 to 12:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "0h15", "dur_3": "-1h30", "dur_0": "0h00", "dur_1": "0h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = $746; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.222", "p2": "0.700", "p0": "0.300", "p1": "0.700"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = 2026-01-12; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 227:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 07:00 to 16:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "8h00", "dur_0": "8h30", "dur_2": "8h45", "dur_1": "9h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 09:15 to 14:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "3h45", "dur_0": "4h30", "dur_2": "5h00", "dur_1": "4h45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = $850; priority = low.\n\nClaim to verify: \"the amount is $850\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = open; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p2": "0.500", "p0": "0.500", "p1": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 168 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "46", "answer_2": "44", "answer_1": "64", "answer_0": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "17", "answer_2": "12", "answer_0": "32", "answer_3": "20"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 62 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "92", "answer_0": "84", "answer_3": "79", "answer_1": "88"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $752. A fee of $55 is charged, then a deposit of $65 arrives. A second fee of $27 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "735", "answer_3": "727", "answer_2": "764", "answer_1": "733"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=653, unit_price=581, discount_pct=0, region=EU, sku_total=1234 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "379393", "v3": "380046", "v2": "37939", "v1": "1234"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1971 per year. It increases by 5% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1872.45", "answer_3": "2044.55", "answer_1": "2069.55", "answer_0": "2025.55"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $489. A fee of $83 is charged, then a deposit of $54 arrives. A second fee of $41 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "409", "answer_3": "422", "answer_1": "432", "answer_0": "419"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.167", "p1": "0.250", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 377:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed March 17, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 24", "date_0": "March 24", "date_2": "March 22", "date_1": "March 22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p0": "0.583", "p2": "0.417", "p3": "0.545"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1965. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "61", "age_3": "62", "age_0": "60", "age_1": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $19k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $763. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=414, unit_price=911, discount_pct=0, region=EU, sku_total=1325 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1325", "v3": "376740", "v2": "37715", "v0": "377154"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=751, unit_price=331, discount_pct=0, region=EU, sku_total=1082 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "24858", "v1": "1082", "v0": "248581", "v3": "247830"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = $273; priority = low.\n\nClaim to verify: \"the owner is $273\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 288 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "96", "answer_0": "80", "answer_3": "100", "answer_1": "63"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = $690; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = $542; priority = low.\n\nClaim to verify: \"the owner is $542\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $202. A fee of $192 is charged, then a deposit of $54 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "-32", "answer_2": "-9", "answer_1": "-61", "answer_3": "-12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p2": "0.125", "p3": "0.125", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = open; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $11k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 335 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "192", "answer_3": "204", "answer_1": "183", "answer_2": "188"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.467", "p0": "0.500", "p1": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 332 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "236", "answer_3": "252", "answer_0": "240", "answer_2": "254"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "8", "answer_0": "32", "answer_1": "14", "answer_3": "24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 348 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "115", "answer_0": "135", "answer_3": "127", "answer_1": "123"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $304. A fee of $159 is charged, then a deposit of $70 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "163", "answer_2": "153", "answer_0": "136", "answer_1": "145"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1950 per year. It increases by 5% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2010.5", "answer_2": "1852.5", "answer_1": "2047.5", "answer_3": "1970.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $485. A fee of $100 is charged, then a deposit of $20 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "329", "answer_1": "360", "answer_2": "378", "answer_0": "355"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = $437; priority = high.\n\nClaim to verify: \"the status is $437\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = 2026-04-17; priority = high.\n\nClaim to verify: \"the owner is 2026-04-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1259 per year. It increases by 25% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1535.75", "answer_1": "1573.75", "answer_2": "944.25", "answer_3": "1566.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1964. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "60", "age_1": "59", "age_2": "61", "age_3": "62"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $67. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p3": "0.050", "p1": "0.200", "p2": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 166 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "79", "answer_2": "45", "answer_0": "63", "answer_1": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 11 people. A shopper bought a 88 gram box and a 880 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 13:30 to 18:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "4h30", "dur_1": "5h15", "dur_3": "4h15", "dur_2": "5h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=361, unit_price=900, discount_pct=0, region=EU, sku_total=1261 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "324900", "v1": "1261", "v2": "32490", "v3": "324539"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 4 grams of salt per serving and serves 11 people. A shopper bought a 44 gram box and a 440 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 768:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1960. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "63", "age_2": "65", "age_3": "62", "age_0": "64"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p2": "0.083", "p1": "0.250", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = $601; priority = low.\n\nClaim to verify: \"the status is $601\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.571", "p2": "0.429", "p3": "0.500", "p1": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $732. A fee of $166 is charged, then a deposit of $71 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "536", "answer_2": "545", "answer_0": "554", "answer_1": "575"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed December 17, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 29", "date_0": "December 31", "date_1": "December 27", "date_2": "December 29"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, June 28, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, February 15, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $559 per year. It increases by 10% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "559.9", "answer_0": "586.9", "answer_1": "614.9", "answer_2": "503.1"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed October 17, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 31", "date_1": "October 26", "date_3": "October 28", "date_0": "October 29"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 296 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "89", "answer_0": "96", "answer_2": "102", "answer_3": "88"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = 2026-03-22; priority = high.\n\nClaim to verify: \"the owner is 2026-03-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "23", "answer_2": "18", "answer_3": "15", "answer_0": "33"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed September 17, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 22", "date_2": "September 22", "date_3": "September 24", "date_0": "September 24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = 2026-12-07; priority = high.\n\nClaim to verify: \"the owner is 2026-12-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $525 per year. It increases by 25% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "641.25", "answer_1": "656.25", "answer_2": "393.75", "answer_3": "603.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $875. A fee of $182 is charged, then a deposit of $76 arrives. A second fee of $91 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "704", "answer_2": "700", "answer_3": "674", "answer_0": "678"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.222", "p1": "0.700", "p0": "0.300", "p2": "0.700"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.538", "p1": "0.462", "p3": "0.500", "p2": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 111 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "280", "answer_1": "283", "answer_2": "287", "answer_3": "270"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 279 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "68", "answer_0": "72", "answer_3": "65", "answer_1": "90"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 11 people. A shopper bought a 33 gram box and a 330 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p0": "0.100", "p2": "0.050", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1966. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "58", "age_1": "59", "age_2": "61", "age_0": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=171, unit_price=831, discount_pct=0, region=EU, sku_total=1002 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "14210", "v1": "1002", "v0": "142101", "v3": "142272"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p0": "0.500", "p3": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 08:15 to 09:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "1h15", "dur_3": "0h15", "dur_0": "0h30", "dur_2": "1h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.538", "p0": "0.571", "p2": "0.429", "p1": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $25k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed January 24, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 29", "date_2": "January 27", "date_3": "January 30", "date_1": "January 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = ops; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = $613; priority = low.\n\nClaim to verify: \"the status is $613\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 263:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 782:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $696. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 504:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, February 16, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = $443; priority = low.\n\nClaim to verify: \"the status is $443\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $236 per year. It increases by 25% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "245", "answer_2": "177", "answer_3": "211", "answer_1": "295"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = 2026-04-18; priority = high.\n\nClaim to verify: \"the amount is 2026-04-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = $962; priority = low.\n\nClaim to verify: \"the status is $962\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1971. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "55", "age_1": "54", "age_3": "57", "age_2": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 7 people. A shopper bought a 49 gram box and a 490 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed February 03, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 09", "date_1": "February 09", "date_3": "February 11", "date_0": "February 11"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 397 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "65", "answer_3": "68", "answer_2": "46", "answer_0": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 385:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 519:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.375", "p0": "0.250", "p2": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.375", "p0": "0.250", "p2": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A recipe needs 8 grams of salt per serving and serves 10 people. A shopper bought a 80 gram box and a 800 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, June 09, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 18:15 to 18:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "0h15", "dur_3": "-1h15", "dur_2": "0h15", "dur_0": "0h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $587 per year. It increases by 10% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "607.7", "answer_1": "645.7", "answer_2": "528.3", "answer_3": "620.7"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 444:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=519, unit_price=865, discount_pct=0, region=EU, sku_total=1384 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "448416", "v2": "44893", "v0": "448935", "v1": "1384"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $479 per year. It increases by 25% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "359.25", "answer_3": "529.75", "answer_0": "548.75", "answer_1": "598.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 9 grams of salt per serving and serves 7 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_3": "32", "answer_0": "48", "answer_1": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $84. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, December 17, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "16", "answer_1": "10", "answer_2": "8", "answer_3": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = $416; priority = low.\n\nClaim to verify: \"the amount is $416\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.714", "p2": "0.286", "p1": "0.286", "p3": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.083", "p1": "0.417", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, December 10, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 480:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 450:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = 2026-05-22; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1965. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "62", "age_1": "60", "age_3": "59", "age_0": "61"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.556", "p1": "0.444", "p3": "0.500", "p2": "0.444"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_3": "32", "answer_1": "52", "answer_2": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 389 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "45", "answer_0": "60", "answer_2": "73", "answer_3": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.333", "p3": "0.625", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "32", "answer_2": "30", "answer_0": "40", "answer_3": "10"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = $127; priority = low.\n\nClaim to verify: \"the owner is $127\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p3": "0.429", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $343. A fee of $130 is charged, then a deposit of $30 arrives. A second fee of $65 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "178", "answer_2": "153", "answer_1": "150", "answer_3": "161"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 843:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 128 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "11", "answer_0": "30", "answer_1": "17", "answer_3": "26"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $321 per year. It increases by 20% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "256.8", "answer_1": "385.2", "answer_0": "342.2", "answer_3": "352.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "62", "age_2": "65", "age_1": "63", "age_0": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $796 per year. It increases by 5% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "756.2", "answer_3": "802.8", "answer_0": "807.8", "answer_1": "835.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.500", "p1": "0.833", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p0": "0.100", "p1": "0.150", "p3": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1960. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "65", "age_3": "66", "age_0": "64", "age_1": "63"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 218 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "135", "answer_2": "152", "answer_0": "144", "answer_3": "132"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1872 per year. It increases by 15% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2163.8", "answer_0": "2139.8", "answer_1": "2152.8", "answer_2": "1591.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p1": "0.615", "p2": "0.615", "p0": "0.385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "35", "answer_3": "35", "answer_1": "40", "answer_0": "70"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 331:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = $791; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=814, unit_price=486, discount_pct=0, region=EU, sku_total=1300 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "395604", "v1": "1300", "v2": "39560", "v3": "394790"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 2 grams of salt per serving and serves 10 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = $794; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $849. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "84", "answer_2": "42", "answer_3": "42", "answer_1": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p2": "0.667", "p0": "0.333", "p1": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = $719; priority = high.\n\nClaim to verify: \"the owner is $719\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, April 12, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "13", "answer_3": "21", "answer_0": "27", "answer_2": "6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $497 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "472.15", "answer_1": "521.85", "answer_0": "489.85", "answer_3": "503.85"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 18:15 to 18:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "0h30", "dur_2": "0h30", "dur_3": "-1h30", "dur_0": "0h00"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A shift runs from 10:30 to 13:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "2h45", "dur_3": "1h45", "dur_1": "2h45", "dur_0": "2h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 880:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 877:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.250", "p3": "0.125", "p0": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = $950; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1961. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "63", "age_0": "65", "age_1": "64", "age_2": "66"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1958. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "65", "age_2": "68", "age_1": "66", "age_0": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = 2026-05-10; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 563:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.150", "p0": "0.100", "p2": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.500", "p0": "0.667", "p1": "0.833"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, March 07, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed August 28, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 02", "date_2": "September 06", "date_0": "September 04", "date_3": "September 04"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $834 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "804.7", "answer_2": "792.3", "answer_0": "843.7", "answer_1": "875.7"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed October 19, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 22", "date_2": "October 24", "date_3": "October 24", "date_0": "October 22"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1428 per year. It increases by 5% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1464.4", "answer_3": "1474.4", "answer_2": "1356.6", "answer_1": "1499.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 10:30 to 16:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "6h00", "dur_3": "5h00", "dur_0": "5h15", "dur_2": "5h45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Friday, April 10, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 131 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "171", "answer_1": "166", "answer_0": "168", "answer_3": "155"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = 2026-05-22; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Monday, July 27, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p3": "0.050", "p0": "0.150", "p2": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 597:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 536:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p1": "0.200", "p3": "0.050", "p2": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=999, unit_price=586, discount_pct=0, region=EU, sku_total=1585 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1585", "v3": "584415", "v2": "58541", "v0": "585414"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed February 24, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 08", "date_1": "March 06", "date_2": "March 08", "date_0": "March 10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "26", "answer_0": "50", "answer_2": "20", "answer_3": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = $327; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1510 per year. It increases by 25% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1887.5", "answer_2": "1132.5", "answer_0": "1868.5", "answer_3": "1843.5"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "18", "answer_0": "30", "answer_3": "15", "answer_2": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "21", "answer_1": "28", "answer_3": "49", "answer_0": "70"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 206:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 8 grams of salt per serving and serves 12 people. A shopper bought a 96 gram box and a 960 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 350 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "115", "answer_2": "83", "answer_0": "96", "answer_1": "113"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 209 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "234", "answer_0": "240", "answer_1": "234", "answer_2": "257"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 08:45 to 13:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "4h15", "dur_3": "3h30", "dur_1": "4h30", "dur_2": "4h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, November 23, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.500", "p3": "0.167", "p1": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 146 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "150", "answer_2": "133", "answer_1": "165", "answer_3": "157"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 451:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = pending; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.083", "p2": "0.250", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = $611; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, June 03, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 240 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "75", "answer_0": "84", "answer_3": "70", "answer_2": "88"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 185 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "227", "answer_3": "244", "answer_0": "225", "answer_2": "209"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1431 per year. It increases by 10% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1570.1", "answer_2": "1287.9", "answer_0": "1549.1", "answer_1": "1574.1"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 55 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "264", "answer_3": "271", "answer_2": "269", "answer_0": "280"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $543 per year. It increases by 25% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "407.25", "answer_3": "676.75", "answer_1": "678.75", "answer_0": "664.75"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 10 people. A shopper bought a 90 gram box and a 900 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = 2026-12-17; priority = low.\n\nClaim to verify: \"the amount is 2026-12-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.600", "p1": "0.400", "p2": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed April 11, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 18", "date_3": "April 20", "date_0": "April 21", "date_2": "April 23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1017 per year. It increases by 20% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "813.6", "answer_1": "1220.4", "answer_3": "1237.4", "answer_0": "1205.4"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1868 per year. It increases by 25% this year, then a $49 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2335", "answer_2": "1401", "answer_3": "2311", "answer_0": "2286"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=810, unit_price=671, discount_pct=0, region=EU, sku_total=1481 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "543510", "v1": "1481", "v2": "54351", "v3": "544320"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = $793; priority = low.\n\nClaim to verify: \"the status is $793\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $30k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $290 per year. It increases by 10% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "302", "answer_3": "286", "answer_1": "319", "answer_2": "261"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 299:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "48", "answer_1": "51", "answer_0": "72", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $828 per year. It increases by 15% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "952.2", "answer_0": "922.2", "answer_2": "703.8", "answer_3": "939.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 14:00 to 16:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "1h45", "dur_3": "1h00", "dur_2": "2h15", "dur_1": "2h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $765. A fee of $165 is charged, then a deposit of $90 arrives. A second fee of $82 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "579", "answer_2": "629", "answer_1": "594", "answer_0": "608"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $16k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = $412; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.727", "p3": "0.200", "p0": "0.273", "p1": "0.727"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $898. A fee of $181 is charged, then a deposit of $36 arrives. A second fee of $90 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "683", "answer_1": "671", "answer_0": "663", "answer_2": "669"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_0": "27", "answer_3": "15", "answer_1": "17"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 77 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "96", "answer_3": "112", "answer_2": "87", "answer_1": "88"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 754:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 384 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "92", "answer_3": "108", "answer_0": "105", "answer_2": "108"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.571", "p3": "0.385", "p1": "0.571", "p0": "0.429"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $342. A fee of $178 is charged, then a deposit of $70 arrives. A second fee of $89 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "173", "answer_3": "158", "answer_0": "145", "answer_2": "140"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = $560; priority = low.\n\nClaim to verify: \"the date is $560\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1733 per year. It increases by 15% this year, then a $40 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1473.05", "answer_1": "1992.95", "answer_3": "1912.95", "answer_0": "1952.95"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = $808; priority = high.\n\nClaim to verify: \"the date is $808\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.250", "p3": "0.083", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 85 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "208", "answer_1": "184", "answer_0": "192", "answer_2": "185"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $784 per year. It increases by 20% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "917.8", "answer_2": "627.2", "answer_1": "940.8", "answer_3": "912.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.150", "p0": "0.100", "p2": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 934:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, June 22, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $887. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "28", "answer_1": "22", "answer_2": "20", "answer_3": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, October 10, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1978. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "45", "age_3": "48", "age_2": "47", "age_0": "46"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1987. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "39", "age_1": "37", "age_0": "38", "age_3": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed November 09, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 19", "date_2": "November 21", "date_3": "November 21", "date_0": "November 23"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 08:15 to 13:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "3h45", "dur_0": "4h00", "dur_2": "4h15", "dur_1": "4h45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $369. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 253:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 8 grams of salt per serving and serves 6 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = 2026-10-11; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $423. A fee of $106 is charged, then a deposit of $78 arrives. A second fee of $53 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "316", "answer_3": "314", "answer_0": "342", "answer_2": "334"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 8 grams of salt per serving and serves 12 people. A shopper bought a 96 gram box and a 960 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=470, unit_price=116, discount_pct=0, region=EU, sku_total=586 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "54520", "v1": "586", "v3": "54990", "v2": "5452"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.333", "p3": "0.167", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p0": "0.500", "p3": "0.125", "p1": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1646 per year. It increases by 15% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1892.9", "answer_0": "1860.9", "answer_2": "1399.1", "answer_3": "1847.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 906:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.400", "p1": "0.400", "p3": "0.500", "p0": "0.600"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 290 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "88", "answer_3": "57", "answer_0": "72", "answer_2": "80"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = open; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1567 per year. It increases by 25% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1175.25", "answer_1": "1958.75", "answer_0": "1920.75", "answer_3": "1914.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 3 grams of salt per serving and serves 5 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p0": "0.500", "p2": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = $732; priority = low.\n\nClaim to verify: \"the owner is $732\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "30", "answer_0": "48", "answer_2": "24", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.167", "p0": "0.333", "p3": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p1": "0.250", "p0": "0.200", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $22k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 296:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed February 01, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 14", "date_3": "February 12", "date_0": "February 12", "date_1": "February 10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed June 15, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 22", "date_0": "June 22", "date_2": "June 24", "date_1": "June 20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = $229; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $19k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.545", "p1": "0.545", "p0": "0.455", "p3": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=698, unit_price=167, discount_pct=0, region=EU, sku_total=865 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "11656", "v1": "865", "v3": "117264", "v0": "116566"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 338 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "123", "answer_1": "110", "answer_0": "108", "answer_2": "106"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1533 per year. It increases by 25% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1848.25", "answer_2": "1149.75", "answer_1": "1916.25", "answer_0": "1874.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=830, unit_price=962, discount_pct=0, region=EU, sku_total=1792 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "797630", "v1": "1792", "v2": "79846", "v0": "798460"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "8", "answer_1": "26", "answer_2": "24", "answer_0": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $275. A fee of $47 is charged, then a deposit of $26 arrives. A second fee of $23 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "231", "answer_1": "214", "answer_2": "223", "answer_3": "220"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1447 per year. It increases by 20% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1706.4", "answer_1": "1736.4", "answer_2": "1157.6", "answer_0": "1722.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $533. A fee of $73 is charged, then a deposit of $37 arrives. A second fee of $36 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "450", "answer_1": "438", "answer_2": "479", "answer_0": "461"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.667", "p3": "0.625", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=977, unit_price=392, discount_pct=0, region=EU, sku_total=1369 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "382984", "v1": "1369", "v3": "383961", "v2": "38298"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed October 08, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 19", "date_3": "October 19", "date_0": "October 21", "date_1": "October 17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 14:30 to 16:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h45", "dur_1": "2h15", "dur_0": "1h30", "dur_3": "1h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.083", "p2": "0.083", "p0": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 796:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.125", "p0": "0.250", "p1": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 18:45 to 21:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h45", "dur_0": "2h15", "dur_3": "1h45", "dur_2": "2h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=277, unit_price=175, discount_pct=0, region=EU, sku_total=452 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "452", "v0": "48475", "v2": "4847", "v3": "48752"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.429", "p1": "0.571", "p3": "0.333", "p2": "0.571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1963. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "60", "age_1": "61", "age_2": "63", "age_0": "62"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 586:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1082 per year. It increases by 10% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1146.2", "answer_1": "1190.2", "answer_2": "973.8", "answer_3": "1121.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 235 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_3": "33", "answer_2": "20", "answer_1": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed April 17, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 20", "date_3": "April 22", "date_0": "April 22", "date_2": "April 24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.429", "p1": "0.533", "p2": "0.533", "p0": "0.467"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, September 04, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = ops; priority = low.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = $681; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=481, unit_price=982, discount_pct=0, region=EU, sku_total=1463 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "472342", "v1": "1463", "v3": "471861", "v2": "47234"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 06:00 to 07:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "0h45", "dur_1": "1h00", "dur_2": "1h00", "dur_3": "0h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 07:15 to 18:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "11h15", "dur_0": "10h45", "dur_3": "10h15", "dur_2": "11h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = $577; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.083", "p2": "0.167", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 15:15 to 21:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "6h00", "dur_2": "6h15", "dur_3": "5h00", "dur_0": "5h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.333", "p0": "0.500", "p1": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = 2026-06-20; priority = low.\n\nClaim to verify: \"the owner is 2026-06-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $346 per year. It increases by 25% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "259.5", "answer_3": "401.5", "answer_0": "406.5", "answer_1": "432.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 06:45 to 14:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "7h45", "dur_1": "7h45", "dur_0": "7h15", "dur_3": "6h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1968. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "57", "age_1": "56", "age_3": "55", "age_2": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed May 01, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 11", "date_0": "May 13", "date_1": "May 09", "date_3": "May 11"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, October 08, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.462", "p1": "0.500", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_1": "37", "answer_2": "35", "answer_0": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 09:45 to 14:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "4h30", "dur_1": "4h45", "dur_0": "4h00", "dur_3": "3h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $31k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $964 per year. It increases by 20% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1156.8", "answer_2": "771.2", "answer_0": "1146.8", "answer_3": "1155.8"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $556. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.467", "p3": "0.429", "p1": "0.533", "p2": "0.533"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 11 people. A shopper bought a 66 gram box and a 660 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = 2026-05-10; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 353 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "84", "answer_2": "70", "answer_1": "74", "answer_0": "72"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=869, unit_price=508, discount_pct=0, region=EU, sku_total=1377 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "442321", "v0": "441452", "v2": "44145", "v1": "1377"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Form fields: status = $67; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.167", "p3": "0.083", "p2": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=566, unit_price=132, discount_pct=0, region=EU, sku_total=698 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "7471", "v1": "698", "v0": "74712", "v3": "75278"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p3": "0.083", "p1": "0.250", "p2": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1968. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "57", "age_2": "58", "age_1": "56", "age_3": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = 2026-11-10; priority = high.\n\nClaim to verify: \"the status is 2026-11-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $812. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1980. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "45", "age_0": "44", "age_1": "43", "age_3": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, March 14, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $319. A fee of $60 is charged, then a deposit of $29 arrives. A second fee of $30 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "258", "answer_3": "233", "answer_1": "261", "answer_2": "278"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $345. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 621:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 245 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "206", "answer_2": "205", "answer_3": "207", "answer_0": "224"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 400:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 333 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "249", "answer_2": "230", "answer_3": "259", "answer_0": "240"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=874, unit_price=145, discount_pct=0, region=EU, sku_total=1019 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1019", "v2": "12673", "v3": "127604", "v0": "126730"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1980. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "45", "age_1": "44", "age_2": "46", "age_3": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.750", "p3": "0.143", "p2": "0.750", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = 2026-11-08; priority = high.\n\nClaim to verify: \"the owner is 2026-11-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $320. A fee of $126 is charged, then a deposit of $53 arrives. A second fee of $63 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "179", "answer_2": "214", "answer_0": "184", "answer_3": "182"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.444", "p2": "0.444", "p3": "0.500", "p0": "0.556"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=572, unit_price=653, discount_pct=0, region=EU, sku_total=1225 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "373516", "v2": "37351", "v1": "1225", "v3": "372944"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = 2026-02-02; priority = low.\n\nClaim to verify: \"the date is 2026-02-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 12 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=439, unit_price=972, discount_pct=0, region=EU, sku_total=1411 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "427147", "v0": "426708", "v1": "1411", "v2": "42670"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "66", "answer_1": "41", "answer_2": "36", "answer_3": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 549:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $499. A fee of $189 is charged, then a deposit of $21 arrives. A second fee of $94 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "237", "answer_1": "261", "answer_3": "266", "answer_2": "225"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $560. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Sunday, June 14, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1547 per year. It increases by 15% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1314.95", "answer_0": "1756.05", "answer_3": "1744.05", "answer_1": "1779.05"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=862, unit_price=556, discount_pct=0, region=EU, sku_total=1418 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "478410", "v2": "47927", "v0": "479272", "v1": "1418"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 242 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "210", "answer_1": "194", "answer_3": "227", "answer_2": "193"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, September 28, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p1": "0.500", "p3": "0.125", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.462", "p0": "0.538", "p3": "0.500", "p2": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed June 04, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 16", "date_1": "June 14", "date_0": "June 18", "date_2": "June 16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p0": "0.100", "p1": "0.150", "p3": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 285 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_3": "48", "answer_2": "17", "answer_1": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1987. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "40", "age_1": "38", "age_3": "41", "age_0": "39"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = $925; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p1": "0.600", "p3": "0.333", "p2": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p1": "0.600", "p2": "0.600", "p0": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $119. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 256:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $379. A fee of $69 is charged, then a deposit of $30 arrives. A second fee of $34 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "316", "answer_2": "290", "answer_1": "314", "answer_0": "306"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = 2026-10-17; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = 2026-06-19; priority = low.\n\nClaim to verify: \"the owner is 2026-06-19\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1987. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "36", "age_3": "35", "age_0": "37", "age_2": "38"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1273 per year. It increases by 15% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1481.95", "answer_1": "1463.95", "answer_2": "1082.05", "answer_0": "1450.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 246 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "260", "answer_1": "253", "answer_2": "237", "answer_0": "240"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1972. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "53", "age_3": "56", "age_0": "54", "age_2": "55"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $5k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p3": "0.167", "p0": "0.500", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $22k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 120 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "128", "answer_2": "140", "answer_1": "113", "answer_3": "139"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed September 23, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 28", "date_3": "September 30", "date_0": "September 30", "date_2": "October 02"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 13, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 21", "date_1": "April 19", "date_0": "April 21", "date_2": "April 19"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $843. A fee of $199 is charged, then a deposit of $38 arrives. A second fee of $99 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "557", "answer_1": "560", "answer_0": "583", "answer_2": "604"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = 2026-03-14; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 855:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = $695; priority = low.\n\nClaim to verify: \"the owner is $695\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 173 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "259", "answer_3": "260", "answer_0": "240", "answer_2": "252"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.538", "p0": "0.462", "p3": "0.417", "p1": "0.538"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $374. A fee of $124 is charged, then a deposit of $72 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "260", "answer_2": "279", "answer_1": "239", "answer_3": "231"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1048 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1100.4", "answer_3": "1028.4", "answer_0": "1068.4", "answer_2": "995.6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, July 28, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p1": "0.833", "p3": "0.167", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed June 04, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 16", "date_0": "June 18", "date_1": "June 14", "date_2": "June 16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.375", "p3": "0.125", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Form fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 60 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "311", "answer_0": "294", "answer_2": "305", "answer_1": "281"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1960. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "65", "age_3": "68", "age_0": "66", "age_2": "67"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $765. A fee of $49 is charged, then a deposit of $28 arrives. A second fee of $24 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "725", "answer_0": "720", "answer_3": "742", "answer_1": "747"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1985. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "37", "age_0": "39", "age_2": "40", "age_1": "38"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 612:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.500", "p2": "0.250", "p0": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 5 grams of salt per serving and serves 5 people. A shopper bought a 25 gram box and a 250 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $12k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Wednesday, May 27, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=871, unit_price=374, discount_pct=0, region=EU, sku_total=1245 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "32575", "v3": "324883", "v1": "1245", "v0": "325754"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.500", "p1": "0.833", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Sunday, December 06, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 951:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $369. A fee of $109 is charged, then a deposit of $56 arrives. A second fee of $54 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "262", "answer_3": "258", "answer_1": "280", "answer_2": "272"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed June 24, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 01", "date_2": "July 05", "date_3": "July 03", "date_0": "July 03"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $306. A fee of $104 is charged, then a deposit of $27 arrives. A second fee of $52 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "198", "answer_2": "157", "answer_1": "164", "answer_0": "177"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=137, unit_price=492, discount_pct=0, region=EU, sku_total=629 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "629", "v0": "67404", "v2": "6740", "v3": "67541"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $403 per year. It increases by 10% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "409.3", "answer_1": "443.3", "answer_2": "362.7", "answer_3": "449.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = $96; priority = low.\n\nClaim to verify: \"the owner is $96\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 904:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = 2026-03-16; priority = high.\n\nClaim to verify: \"the status is 2026-03-16\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = 2026-08-14; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 09:45 to 10:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "0h00", "dur_1": "0h30", "dur_3": "-1h30", "dur_2": "0h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $351. A fee of $53 is charged, then a deposit of $30 arrives. A second fee of $26 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "302", "answer_1": "321", "answer_3": "324", "answer_2": "275"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 12:45 to 20:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "6h45", "dur_3": "6h15", "dur_0": "6h30", "dur_1": "7h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = 2026-05-01; priority = low.\n\nClaim to verify: \"the date is 2026-05-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 589:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p1": "0.150", "p0": "0.100", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.286", "p2": "0.625", "p1": "0.625", "p0": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed June 07, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 16", "date_3": "June 18", "date_0": "June 18", "date_1": "June 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=602, unit_price=478, discount_pct=0, region=EU, sku_total=1080 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "287154", "v2": "28775", "v1": "1080", "v0": "287756"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $234. A fee of $144 is charged, then a deposit of $68 arrives. A second fee of $72 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "64", "answer_1": "80", "answer_0": "86", "answer_3": "101"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_1": "19", "answer_2": "12", "answer_0": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p2": "0.571", "p0": "0.429", "p3": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = $317; priority = high.\n\nClaim to verify: \"the owner is $317\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = $536; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1964. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "61", "age_3": "62", "age_0": "60", "age_1": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 465:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.400", "p0": "0.455", "p1": "0.545", "p2": "0.545"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1962. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "60", "age_0": "62", "age_1": "61", "age_2": "63"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 185 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "110", "answer_1": "107", "answer_2": "88", "answer_0": "90"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1964. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "63", "age_0": "62", "age_3": "60", "age_1": "61"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 232 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "73", "answer_0": "56", "answer_2": "68", "answer_3": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 110 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "84", "answer_2": "66", "answer_0": "81", "answer_3": "67"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=442, unit_price=968, discount_pct=0, region=EU, sku_total=1410 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1410", "v0": "427856", "v3": "427414", "v2": "42785"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, December 09, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A shift runs from 13:45 to 16:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h15", "dur_1": "2h30", "dur_2": "2h30", "dur_3": "1h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 288 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "192", "answer_1": "177", "answer_2": "189", "answer_3": "178"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p3": "0.444", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.467", "p0": "0.533", "p2": "0.467"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.167", "p0": "0.333", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 829:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $406. A fee of $137 is charged, then a deposit of $38 arrives. A second fee of $68 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "223", "answer_2": "249", "answer_0": "239", "answer_3": "258"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, March 20, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, May 06, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.571", "p2": "0.429", "p3": "0.500", "p1": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "42", "answer_1": "49", "answer_0": "91", "answer_3": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 216 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "121", "answer_0": "105", "answer_3": "98", "answer_2": "114"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 113 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "276", "answer_3": "273", "answer_2": "275", "answer_0": "270"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $612 per year. It increases by 5% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "581.4", "answer_3": "609.6", "answer_0": "600.6", "answer_1": "642.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $436. A fee of $191 is charged, then a deposit of $59 arrives. A second fee of $95 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "181", "answer_0": "209", "answer_2": "225", "answer_1": "225"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 782:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=770, unit_price=675, discount_pct=0, region=EU, sku_total=1445 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1445", "v0": "519750", "v3": "518980", "v2": "51975"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $351. A fee of $179 is charged, then a deposit of $42 arrives. A second fee of $89 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "150", "answer_1": "114", "answer_0": "125", "answer_3": "116"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = 2026-08-20; priority = low.\n\nClaim to verify: \"the amount is 2026-08-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.417", "p3": "0.364", "p1": "0.583", "p2": "0.583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 397 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "140", "answer_3": "143", "answer_0": "126", "answer_1": "130"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 15:45 to 17:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "2h15", "dur_0": "1h45", "dur_3": "1h00", "dur_1": "2h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $743. A fee of $113 is charged, then a deposit of $56 arrives. A second fee of $56 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "650", "answer_2": "635", "answer_0": "630", "answer_1": "637"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = $260; priority = high.\n\nClaim to verify: \"the date is $260\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Sunday, March 15, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.600", "p1": "0.400", "p2": "0.400", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.538", "p3": "0.500", "p2": "0.462", "p1": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "31", "answer_0": "66", "answer_3": "42", "answer_2": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed September 10, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 25", "date_0": "September 23", "date_1": "September 19", "date_3": "September 21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $645. A fee of $144 is charged, then a deposit of $41 arrives. A second fee of $72 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "470", "answer_3": "464", "answer_2": "473", "answer_1": "499"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.200", "p0": "0.333", "p1": "0.667", "p2": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.385", "p2": "0.571", "p1": "0.571", "p0": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $342 per year. It increases by 15% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "290.7", "answer_3": "410.3", "answer_1": "393.3", "answer_0": "376.3"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 17:00 to 17:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "0h45", "dur_2": "0h15", "dur_0": "0h00", "dur_3": "-1h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 391 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "105", "answer_1": "98", "answer_0": "96", "answer_3": "76"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $693. A fee of $62 is charged, then a deposit of $36 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "644", "answer_1": "628", "answer_3": "639", "answer_0": "636"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $246. A fee of $156 is charged, then a deposit of $81 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "93", "answer_3": "122", "answer_1": "71", "answer_2": "78"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = $951; priority = low.\n\nClaim to verify: \"the amount is $951\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed February 13, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 27", "date_1": "February 21", "date_0": "February 25", "date_3": "February 23"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1966. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "57", "age_2": "59", "age_0": "58", "age_3": "56"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 652:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $39k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Thursday, February 26, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = $222; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1967. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "60", "age_0": "59", "age_1": "58", "age_3": "61"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "25", "answer_3": "15", "answer_2": "10", "answer_1": "13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "32", "answer_1": "20", "answer_0": "48", "answer_2": "16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = 2026-04-11; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 328 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "69", "answer_0": "72", "answer_1": "77", "answer_3": "60"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.667", "p3": "0.250", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1970. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "56", "age_3": "54", "age_1": "55", "age_2": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed May 27, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 07", "date_2": "June 07", "date_1": "June 05", "date_0": "June 09"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.727", "p0": "0.273", "p2": "0.727", "p3": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = 2026-12-28; priority = high.\n\nClaim to verify: \"the owner is 2026-12-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.667", "p2": "0.333", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1980. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "44", "age_2": "46", "age_0": "45", "age_3": "43"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = 2026-09-22; priority = high.\n\nClaim to verify: \"the amount is 2026-09-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "38", "answer_2": "35", "answer_3": "21", "answer_0": "56"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = 2026-02-20; priority = low.\n\nClaim to verify: \"the status is 2026-02-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = $658; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.364", "p1": "0.364", "p0": "0.636", "p3": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.273", "p2": "0.273", "p3": "0.700", "p0": "0.727"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, January 03, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $480. A fee of $104 is charged, then a deposit of $25 arrives. A second fee of $52 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "349", "answer_1": "330", "answer_3": "346", "answer_2": "330"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Saturday, July 11, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 06:15 to 17:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "10h00", "dur_1": "10h45", "dur_2": "10h15", "dur_3": "9h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $401. A fee of $48 is charged, then a deposit of $81 arrives. A second fee of $24 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "400", "answer_0": "410", "answer_3": "437", "answer_1": "428"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 09:00 to 13:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "4h00", "dur_0": "3h30", "dur_3": "3h00", "dur_2": "3h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = $899; priority = low.\n\nClaim to verify: \"the amount is $899\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1407 per year. It increases by 15% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1602.05", "answer_3": "1612.05", "answer_1": "1618.05", "answer_2": "1195.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=202, unit_price=796, discount_pct=0, region=EU, sku_total=998 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "160792", "v1": "998", "v2": "16079", "v3": "160994"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = $97; priority = low.\n\nClaim to verify: \"the amount is $97\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 3 grams of salt per serving and serves 12 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.444", "p0": "0.556", "p2": "0.444", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $523. A fee of $124 is charged, then a deposit of $62 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "374", "answer_0": "399", "answer_1": "386", "answer_3": "390"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p2": "0.150", "p1": "0.250", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, February 24, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = 2026-01-08; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, November 14, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = $991; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1679 per year. It increases by 5% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1762.95", "answer_0": "1740.95", "answer_3": "1761.95", "answer_2": "1595.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 624:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $700. A fee of $167 is charged, then a deposit of $47 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "491", "answer_1": "476", "answer_2": "509", "answer_0": "497"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $19k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 315 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "238", "answer_1": "260", "answer_0": "252", "answer_2": "258"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p3": "0.167", "p0": "0.500", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 567:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, August 07, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $227. A fee of $72 is charged, then a deposit of $72 arrives. A second fee of $36 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "189", "answer_0": "191", "answer_1": "175", "answer_3": "199"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed March 18, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 27", "date_0": "March 25", "date_3": "March 25", "date_1": "March 23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = $186; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $347. A fee of $50 is charged, then a deposit of $24 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "319", "answer_0": "296", "answer_2": "315", "answer_1": "289"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1989. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "37", "age_2": "36", "age_1": "34", "age_0": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = open; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1971. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "53", "age_0": "54", "age_2": "55", "age_3": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1976. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "49", "age_2": "50", "age_3": "51", "age_1": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 997:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 5 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, April 08, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $731. A fee of $68 is charged, then a deposit of $32 arrives. A second fee of $34 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "661", "answer_3": "679", "answer_1": "639", "answer_2": "670"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=304, unit_price=964, discount_pct=0, region=EU, sku_total=1268 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "293056", "v3": "293360", "v2": "29305", "v1": "1268"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $857 per year. It increases by 25% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1057.25", "answer_3": "1044.25", "answer_2": "642.75", "answer_1": "1071.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $27k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $681. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.556", "p0": "0.444", "p1": "0.556", "p3": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 424:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed December 03, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 09", "date_0": "December 09", "date_2": "December 11", "date_1": "December 07"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed February 06, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 09", "date_2": "February 09", "date_3": "February 11", "date_0": "February 11"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $265. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $355 per year. It increases by 25% this year, then a $40 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "266.25", "answer_1": "443.75", "answer_0": "403.75", "answer_3": "367.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = $733; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = 2026-08-28; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 620:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, March 10, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed October 10, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "October 14", "date_1": "October 13", "date_3": "October 15", "date_2": "October 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $680. A fee of $102 is charged, then a deposit of $75 arrives. A second fee of $51 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "602", "answer_1": "619", "answer_2": "588", "answer_3": "612"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=879, unit_price=842, discount_pct=0, region=EU, sku_total=1721 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "74011", "v1": "1721", "v0": "740118", "v3": "740997"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 12:15 to 14:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "2h45", "dur_1": "2h30", "dur_3": "1h30", "dur_0": "2h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 229 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "53", "answer_0": "72", "answer_1": "69", "answer_2": "83"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_1": "22", "answer_2": "16", "answer_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.250", "p1": "0.417", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 614:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1362 per year. It increases by 15% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1566.3", "answer_3": "1548.3", "answer_0": "1528.3", "answer_2": "1157.7"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1982. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "44", "age_2": "43", "age_1": "41", "age_0": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 10:15 to 18:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "7h45", "dur_2": "7h45", "dur_3": "6h45", "dur_0": "7h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "14", "answer_0": "18", "answer_3": "6", "answer_2": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = closed; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Monday, May 25, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1964. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "59", "age_2": "61", "age_0": "60", "age_3": "62"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 853:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed January 21, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 28", "date_3": "January 30", "date_0": "January 30", "date_1": "January 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed October 11, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 18", "date_0": "October 20", "date_3": "October 20", "date_1": "October 18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed July 07, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 14", "date_1": "July 14", "date_0": "July 16", "date_3": "July 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $576. A fee of $97 is charged, then a deposit of $75 arrives. A second fee of $48 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "506", "answer_1": "527", "answer_2": "492", "answer_3": "487"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1165 per year. It increases by 15% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1295.75", "answer_1": "1339.75", "answer_0": "1305.75", "answer_2": "990.25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $667. A fee of $47 is charged, then a deposit of $33 arrives. A second fee of $23 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "602", "answer_0": "630", "answer_3": "638", "answer_1": "613"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 125 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "54", "answer_2": "70", "answer_3": "36", "answer_1": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 776:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "51", "age_0": "52", "age_2": "53", "age_3": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p2": "0.600", "p0": "0.400", "p3": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 345:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_2": "14", "answer_0": "28", "answer_1": "16"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 12:30 to 16:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "3h30", "dur_3": "2h30", "dur_2": "3h45", "dur_0": "3h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed May 07, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 13", "date_3": "May 13", "date_2": "May 11", "date_1": "May 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = 2026-10-14; priority = high.\n\nClaim to verify: \"the owner is 2026-10-14\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 169 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_1": "21", "answer_2": "47", "answer_3": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $441. A fee of $30 is charged, then a deposit of $76 arrives. A second fee of $15 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "468", "answer_3": "501", "answer_1": "492", "answer_0": "472"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, July 09, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 660:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = 2026-08-24; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Friday, July 03, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 4 grams of salt per serving and serves 11 people. A shopper bought a 44 gram box and a 440 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $8k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "20", "answer_0": "30", "answer_1": "22", "answer_3": "10"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.222", "p3": "0.125", "p1": "0.778", "p2": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 3 grams of salt per serving and serves 9 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 853:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.300", "p3": "0.222", "p1": "0.700", "p2": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1985. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "41", "age_1": "39", "age_3": "42", "age_0": "40"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $408 per year. It increases by 20% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "489.6", "answer_2": "326.4", "answer_0": "469.6", "answer_3": "436.6"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 317 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "270", "answer_3": "299", "answer_0": "280", "answer_2": "296"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 06:45 to 15:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "7h45", "dur_0": "8h00", "dur_1": "8h45", "dur_2": "8h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = pending; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1989. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "38", "age_1": "35", "age_0": "36", "age_2": "37"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 678:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = 2026-02-18; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1938 per year. It increases by 20% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2291.6", "answer_1": "2325.6", "answer_3": "2305.6", "answer_2": "1550.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p2": "0.100", "p3": "0.050", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, April 24, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_1": "24", "answer_0": "36", "answer_2": "18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 841:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1981. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "41", "age_1": "42", "age_0": "43", "age_2": "44"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1725 per year. It increases by 20% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2070", "answer_2": "1380", "answer_0": "2022", "answer_3": "2010"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = $958; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = closed; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed August 28, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 01", "date_1": "September 01", "date_3": "September 03", "date_0": "September 03"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = 2026-02-14; priority = high.\n\nClaim to verify: \"the owner is 2026-02-14\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p1": "0.500", "p3": "0.125", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed June 02, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "June 12", "date_3": "June 12", "date_1": "June 10", "date_2": "June 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=957, unit_price=292, discount_pct=0, region=EU, sku_total=1249 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "279444", "v1": "1249", "v2": "27944", "v3": "278487"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = closed; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, June 23, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 650:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = $704; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, April 24, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed July 22, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "August 03", "date_2": "August 05", "date_3": "August 01", "date_1": "July 30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.533", "p3": "0.429", "p1": "0.533", "p0": "0.467"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "15", "answer_3": "6", "answer_2": "9", "answer_1": "11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 230:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 270 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "219", "answer_1": "228", "answer_2": "230", "answer_0": "216"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, January 12, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed May 16, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 27", "date_2": "May 29", "date_3": "May 26", "date_1": "May 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 149 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "128", "answer_3": "137", "answer_1": "124", "answer_0": "126"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $423. A fee of $39 is charged, then a deposit of $29 arrives. A second fee of $19 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "409", "answer_2": "373", "answer_0": "394", "answer_3": "372"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = 2026-01-04; priority = high.\n\nClaim to verify: \"the amount is 2026-01-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = ops; priority = low.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = $192; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.700", "p2": "0.273", "p0": "0.727", "p1": "0.273"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 3 grams of salt per serving and serves 9 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 21:30 to 21:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h15", "dur_1": "0h15", "dur_2": "0h30", "dur_0": "0h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 584:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = open; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $771. A fee of $140 is charged, then a deposit of $64 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "649", "answer_3": "607", "answer_0": "625", "answer_1": "619"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, January 27, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = $235; priority = low.\n\nClaim to verify: \"the owner is $235\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=221, unit_price=923, discount_pct=0, region=EU, sku_total=1144 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "20398", "v0": "203983", "v1": "1144", "v3": "203762"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=669, unit_price=919, discount_pct=0, region=EU, sku_total=1588 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "61481", "v3": "615480", "v1": "1588", "v0": "614811"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = ops; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $39k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1979. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "45", "age_2": "47", "age_3": "48", "age_0": "46"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=688, unit_price=548, discount_pct=0, region=EU, sku_total=1236 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "377024", "v2": "37702", "v3": "377712", "v1": "1236"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = 2026-07-22; priority = low.\n\nClaim to verify: \"the amount is 2026-07-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = 2026-10-16; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.273", "p0": "0.727", "p3": "0.700", "p1": "0.273"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.667", "p3": "0.167", "p1": "0.833"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = $692; priority = high.\n\nClaim to verify: \"the owner is $692\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 744:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "30", "answer_3": "20", "answer_1": "14", "answer_2": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $851. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $558. A fee of $171 is charged, then a deposit of $23 arrives. A second fee of $85 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "332", "answer_3": "305", "answer_0": "325", "answer_2": "301"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.571", "p2": "0.429", "p3": "0.538", "p1": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = $237; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 55 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "76", "answer_0": "90", "answer_2": "86", "answer_1": "70"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = $717; priority = low.\n\nClaim to verify: \"the owner is $717\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed October 09, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "October 16", "date_3": "October 16", "date_2": "October 18", "date_1": "October 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.273", "p1": "0.273", "p3": "0.700", "p0": "0.727"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "25", "answer_1": "17", "answer_2": "15", "answer_3": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Friday, March 20, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_0": "24", "answer_2": "12", "answer_1": "14"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = 2026-05-05; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1968. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "59", "age_0": "58", "age_3": "56", "age_1": "57"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p3": "0.050", "p2": "0.050", "p1": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, March 23, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.667", "p0": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.556", "p2": "0.444", "p1": "0.444"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "6", "answer_0": "21", "answer_1": "11", "answer_3": "15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1976. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "48", "age_3": "51", "age_2": "50", "age_0": "49"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $533. A fee of $72 is charged, then a deposit of $70 arrives. A second fee of $36 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "495", "answer_3": "505", "answer_1": "474", "answer_2": "512"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1001 per year. It increases by 5% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1018.05", "answer_1": "1051.05", "answer_2": "950.95", "answer_3": "1027.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 07:15 to 09:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "1h30", "dur_2": "2h45", "dur_1": "2h30", "dur_0": "2h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: owner = $155; priority = high.\n\nClaim to verify: \"the owner is $155\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1634 per year. It increases by 15% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1388.9", "answer_3": "1833.1", "answer_1": "1879.1", "answer_0": "1858.1"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 07:15 to 14:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "6h00", "dur_1": "6h45", "dur_2": "6h15", "dur_3": "5h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $735. A fee of $87 is charged, then a deposit of $64 arrives. A second fee of $43 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "682", "answer_0": "669", "answer_3": "641", "answer_2": "641"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $888. A fee of $98 is charged, then a deposit of $35 arrives. A second fee of $49 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "801", "answer_0": "776", "answer_3": "792", "answer_1": "746"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $295. A fee of $122 is charged, then a deposit of $41 arrives. A second fee of $61 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "160", "answer_1": "161", "answer_0": "153", "answer_2": "175"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 360 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_3": "14", "answer_0": "12", "answer_1": "-8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1985. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "39", "age_0": "40", "age_3": "38", "age_2": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = $461; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, September 14, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 397 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "98", "answer_2": "102", "answer_0": "90", "answer_3": "94"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1958. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "65", "age_1": "66", "age_0": "67", "age_2": "68"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = $942; priority = low.\n\nClaim to verify: \"the status is $942\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "14", "answer_0": "18", "answer_3": "6", "answer_2": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 876:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1967. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "58", "age_0": "57", "age_3": "55", "age_1": "56"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $40k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed October 26, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 06", "date_1": "November 02", "date_0": "November 04", "date_3": "November 04"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, January 26, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "22", "answer_2": "20", "answer_3": "8", "answer_0": "28"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 13:30 to 17:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "4h00", "dur_3": "3h15", "dur_1": "4h15", "dur_0": "3h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1515 per year. It increases by 15% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1752.25", "answer_0": "1727.25", "answer_2": "1287.75", "answer_1": "1742.25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, May 13, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $321. A fee of $136 is charged, then a deposit of $90 arrives. A second fee of $68 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "207", "answer_2": "185", "answer_1": "219", "answer_3": "203"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.571", "p2": "0.429", "p1": "0.429", "p3": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = $119; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_3": "10", "answer_0": "40", "answer_1": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = ops; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed April 10, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 13", "date_2": "April 17", "date_0": "April 15", "date_3": "April 15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 2 grams of salt per serving and serves 8 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1989. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "36", "age_0": "37", "age_3": "39", "age_2": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 506:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1964. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "62", "age_3": "59", "age_1": "60", "age_0": "61"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 11 people. A shopper bought a 66 gram box and a 660 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 938:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $316. A fee of $130 is charged, then a deposit of $60 arrives. A second fee of $65 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "170", "answer_0": "181", "answer_1": "157", "answer_3": "172"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p3": "0.125", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p3": "0.200", "p2": "0.667", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = 2026-02-23; priority = low.\n\nClaim to verify: \"the status is 2026-02-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = 2026-02-05; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $209. A fee of $57 is charged, then a deposit of $69 arrives. A second fee of $28 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "180", "answer_1": "171", "answer_0": "193", "answer_3": "221"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $331. A fee of $189 is charged, then a deposit of $47 arrives. A second fee of $94 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "67", "answer_1": "117", "answer_2": "108", "answer_0": "95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed October 01, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 12", "date_0": "October 14", "date_2": "October 12", "date_1": "October 10"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 271 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "44", "answer_2": "35", "answer_0": "24", "answer_1": "29"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 3 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $802. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed August 20, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 05", "date_0": "September 03", "date_1": "August 30", "date_3": "September 01"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = 2026-09-08; priority = high.\n\nClaim to verify: \"the date is 2026-09-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = $702; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.200", "p3": "0.050", "p2": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = 2026-04-23; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.500", "p1": "0.667", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.714", "p1": "0.250", "p0": "0.750", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Sunday, October 11, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = ops; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 289 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "8", "answer_1": "10", "answer_3": "38", "answer_0": "24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $202 per year. It increases by 25% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "252.5", "answer_2": "151.5", "answer_0": "223.5", "answer_3": "210.5"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 887:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 6 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 823:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $508. A fee of $81 is charged, then a deposit of $87 arrives. A second fee of $40 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "474", "answer_1": "482", "answer_3": "504", "answer_2": "488"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $35k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 13:45 to 22:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "8h30", "dur_0": "8h00", "dur_3": "7h30", "dur_2": "8h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 5 grams of salt per serving and serves 9 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_2": "16", "answer_1": "21", "answer_3": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=867, unit_price=219, discount_pct=0, region=EU, sku_total=1086 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "18987", "v0": "189873", "v3": "190740", "v1": "1086"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $20k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 19:45 to 22:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h30", "dur_3": "1h30", "dur_2": "2h15", "dur_0": "2h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = 2026-09-12; priority = high.\n\nClaim to verify: \"the amount is 2026-09-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 601:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 804:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 349 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "277", "answer_0": "288", "answer_3": "304", "answer_2": "302"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed July 20, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 25", "date_1": "July 23", "date_0": "July 23", "date_2": "July 25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 392:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.800", "p2": "0.800", "p3": "0.111", "p0": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = ops; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $839 per year. It increases by 10% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "884.9", "answer_1": "922.9", "answer_3": "906.9", "answer_2": "755.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=217, unit_price=708, discount_pct=0, region=EU, sku_total=925 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "153853", "v2": "15363", "v0": "153636", "v1": "925"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.125", "p0": "0.375", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 388 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "128", "answer_3": "137", "answer_2": "136", "answer_1": "143"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = $193; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: status = $844; priority = low.\n\nClaim to verify: \"the status is $844\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1959. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "67", "age_2": "66", "age_0": "65", "age_1": "64"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 10:00 to 20:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "10h30", "dur_0": "10h00", "dur_1": "10h45", "dur_3": "9h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $43; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 439:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, December 26, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.375", "p2": "0.250", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = pending; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 852:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.556", "p1": "0.556", "p0": "0.444", "p3": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $695. A fee of $51 is charged, then a deposit of $70 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "716", "answer_0": "689", "answer_2": "669", "answer_1": "687"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Sunday, July 19, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1967 per year. It increases by 25% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2427.75", "answer_3": "2434.75", "answer_2": "1475.25", "answer_1": "2458.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 806:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "10", "answer_0": "30", "answer_1": "22", "answer_2": "20"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p3": "0.050", "p2": "0.050", "p0": "0.100"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, January 10, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.462", "p2": "0.538", "p3": "0.417", "p1": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = 2026-04-24; priority = low.\n\nClaim to verify: \"the owner is 2026-04-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: status = $528; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p1": "0.625", "p0": "0.500", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 167 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "70", "answer_1": "80", "answer_3": "66", "answer_2": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $385 per year. It increases by 25% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "454.25", "answer_3": "483.25", "answer_1": "481.25", "answer_2": "288.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 275 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "212", "answer_1": "172", "answer_0": "192", "answer_3": "182"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 18:45 to 19:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h15", "dur_1": "1h00", "dur_3": "0h00", "dur_0": "0h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = $557; priority = low.\n\nClaim to verify: \"the amount is $557\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p1": "0.250", "p3": "0.083", "p2": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "20", "answer_0": "32", "answer_3": "12", "answer_1": "23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 10 people. A shopper bought a 50 gram box and a 500 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = open; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed July 10, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 18", "date_0": "July 20", "date_3": "July 18", "date_1": "July 16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 07:00 to 21:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "13h15", "dur_1": "14h15", "dur_2": "14h15", "dur_0": "13h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = 2026-06-01; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = 2026-11-01; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: owner = 2026-04-07; priority = low.\n\nClaim to verify: \"the owner is 2026-04-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = 2026-10-03; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.375", "p2": "0.125", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = 2026-08-24; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, April 11, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.429", "p0": "0.571", "p2": "0.429", "p3": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 2 grams of salt per serving and serves 7 people. A shopper bought a 14 gram box and a 140 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 393 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "52", "answer_3": "88", "answer_0": "72", "answer_1": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $416. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 7 grams of salt per serving and serves 12 people. A shopper bought a 84 gram box and a 840 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_3": "30", "answer_2": "18", "answer_1": "23"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 14:30 to 15:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "0h45", "dur_1": "1h15", "dur_2": "1h00", "dur_3": "0h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = $113; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=612, unit_price=333, discount_pct=0, region=EU, sku_total=945 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "203796", "v3": "204408", "v2": "20379", "v1": "945"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $252. A fee of $139 is charged, then a deposit of $37 arrives. A second fee of $69 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "83", "answer_0": "81", "answer_3": "79", "answer_2": "102"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.385", "p0": "0.429", "p1": "0.571", "p2": "0.571"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_0": "40", "answer_1": "22", "answer_3": "24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 534:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.533", "p2": "0.533", "p0": "0.467", "p3": "0.429"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = $531; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 550:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed April 23, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 26", "date_3": "April 28", "date_0": "April 28", "date_2": "April 30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.364", "p0": "0.636", "p2": "0.364", "p3": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 17:15 to 21:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "2h45", "dur_0": "3h30", "dur_2": "4h00", "dur_1": "3h45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "38", "answer_3": "48", "answer_0": "80", "answer_2": "32"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.778", "p0": "0.222", "p3": "0.125", "p2": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = $476; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "63", "answer_2": "42", "answer_1": "45", "answer_3": "21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $826. A fee of $35 is charged, then a deposit of $39 arrives. A second fee of $17 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "784", "answer_3": "842", "answer_2": "828", "answer_0": "813"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.667", "p0": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1493 per year. It increases by 20% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1791.6", "answer_2": "1194.4", "answer_0": "1753.6", "answer_3": "1717.6"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=636, unit_price=587, discount_pct=0, region=EU, sku_total=1223 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "372696", "v0": "373332", "v1": "1223", "v2": "37333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1967 per year. It increases by 5% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2032.35", "answer_2": "1868.65", "answer_3": "2052.35", "answer_1": "2065.35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $404 per year. It increases by 25% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "303", "answer_0": "468", "answer_1": "505", "answer_3": "507"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $383. A fee of $109 is charged, then a deposit of $60 arrives. A second fee of $54 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "297", "answer_0": "280", "answer_2": "257", "answer_3": "306"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $620. A fee of $198 is charged, then a deposit of $29 arrives. A second fee of $99 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "328", "answer_2": "366", "answer_0": "352", "answer_3": "367"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.167", "p1": "0.333", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $426. A fee of $88 is charged, then a deposit of $31 arrives. A second fee of $44 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "338", "answer_0": "325", "answer_1": "311", "answer_2": "296"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 4 people. A shopper bought a 8 gram box and a 80 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_1": "26", "answer_0": "32", "answer_3": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.250", "p0": "0.167", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1812 per year. It increases by 5% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1875.6", "answer_3": "1862.6", "answer_1": "1902.6", "answer_2": "1721.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, November 14, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed February 22, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 28", "date_0": "March 02", "date_2": "February 28", "date_3": "March 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "55", "answer_1": "35", "answer_3": "25", "answer_2": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 253 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "84", "answer_3": "73", "answer_0": "75", "answer_1": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $498. A fee of $68 is charged, then a deposit of $20 arrives. A second fee of $34 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "387", "answer_1": "406", "answer_0": "416", "answer_2": "391"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, February 23, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = $843; priority = high.\n\nClaim to verify: \"the status is $843\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "30", "answer_0": "60", "answer_2": "30", "answer_1": "35"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = 2026-01-23; priority = high.\n\nClaim to verify: \"the amount is 2026-01-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.286", "p0": "0.375", "p1": "0.625", "p2": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "63", "answer_1": "45", "answer_2": "42", "answer_3": "21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A shift runs from 06:15 to 17:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "10h45", "dur_1": "11h15", "dur_2": "11h00", "dur_3": "10h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Friday, April 10, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.778", "p0": "0.800", "p2": "0.200", "p1": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p1": "0.800", "p2": "0.800", "p3": "0.111"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = $401; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = 2026-01-07; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=264, unit_price=372, discount_pct=0, region=EU, sku_total=636 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "636", "v0": "98208", "v2": "9820", "v3": "98472"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 764:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 09:15 to 10:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "0h15", "dur_0": "0h45", "dur_2": "1h00", "dur_1": "1h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 813:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p3": "0.083", "p0": "0.167", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.667", "p3": "0.167", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_3": "40", "answer_0": "56", "answer_1": "21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = $559; priority = high.\n\nClaim to verify: \"the status is $559\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 168 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "89", "answer_3": "62", "answer_2": "67", "answer_0": "75"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1251 per year. It increases by 15% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1421.65", "answer_1": "1438.65", "answer_2": "1063.35", "answer_3": "1457.65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = 2026-12-03; priority = low.\n\nClaim to verify: \"the owner is 2026-12-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 08:30 to 21:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "12h45", "dur_3": "11h45", "dur_0": "12h30", "dur_2": "13h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1431 per year. It increases by 15% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1645.65", "answer_2": "1216.35", "answer_0": "1615.65", "answer_3": "1631.65"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 247 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "56", "answer_1": "80", "answer_0": "64", "answer_3": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.667", "p2": "0.300", "p0": "0.700", "p1": "0.300"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, October 19, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 185 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "58", "answer_0": "45", "answer_2": "35", "answer_3": "26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.500", "p3": "0.167", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $35k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed January 05, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 09", "date_2": "January 07", "date_3": "January 11", "date_0": "January 09"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed March 06, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 14", "date_1": "March 12", "date_0": "March 16", "date_2": "March 14"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 298:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1390 per year. It increases by 5% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1424.5", "answer_3": "1394.5", "answer_1": "1459.5", "answer_2": "1320.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p2": "0.100", "p1": "0.200", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.300", "p3": "0.222", "p1": "0.700", "p2": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1317 per year. It increases by 20% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1569.4", "answer_0": "1552.4", "answer_2": "1053.6", "answer_1": "1580.4"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = 2026-10-16; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.615", "p2": "0.615", "p0": "0.385", "p3": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = 2026-03-27; priority = low.\n\nClaim to verify: \"the amount is 2026-03-27\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $627 per year. It increases by 5% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "625.35", "answer_3": "640.35", "answer_1": "658.35", "answer_2": "595.65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, April 23, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1984. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "41", "age_2": "43", "age_3": "40", "age_0": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.125", "p1": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=820, unit_price=519, discount_pct=0, region=EU, sku_total=1339 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "42558", "v0": "425580", "v1": "1339", "v3": "424760"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1988. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "37", "age_3": "36", "age_0": "38", "age_2": "39"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 84 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "112", "answer_3": "94", "answer_1": "105", "answer_2": "132"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p1": "0.200", "p3": "0.050", "p2": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p1": "0.150", "p3": "0.050", "p2": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 133 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "88", "answer_0": "72", "answer_2": "79", "answer_3": "90"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $601. A fee of $50 is charged, then a deposit of $22 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "555", "answer_0": "548", "answer_2": "558", "answer_1": "522"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $598. A fee of $161 is charged, then a deposit of $71 arrives. A second fee of $80 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "428", "answer_1": "452", "answer_2": "406", "answer_3": "425"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.250", "p2": "0.667", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Sunday, November 08, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=767, unit_price=281, discount_pct=0, region=EU, sku_total=1048 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "216294", "v2": "21552", "v0": "215527", "v1": "1048"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $207 per year. It increases by 5% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "196.65", "answer_3": "194.35", "answer_1": "217.35", "answer_0": "183.35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=644, unit_price=219, discount_pct=0, region=EU, sku_total=863 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "863", "v2": "14103", "v3": "140392", "v0": "141036"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 383:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "28", "answer_0": "56", "answer_3": "28", "answer_1": "32"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=691, unit_price=430, discount_pct=0, region=EU, sku_total=1121 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "297130", "v2": "29713", "v3": "297821", "v1": "1121"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed July 14, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 20", "date_3": "July 22", "date_0": "July 22", "date_2": "July 24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $468 per year. It increases by 25% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "565", "answer_3": "552", "answer_2": "351", "answer_1": "585"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=629, unit_price=428, discount_pct=0, region=EU, sku_total=1057 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "269841", "v1": "1057", "v0": "269212", "v2": "26921"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=384, unit_price=764, discount_pct=0, region=EU, sku_total=1148 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "29337", "v1": "1148", "v3": "293760", "v0": "293376"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 2 grams of salt per serving and serves 4 people. A shopper bought a 8 gram box and a 80 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 633:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_0": "32", "answer_1": "14", "answer_2": "8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, September 11, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = $29; priority = low.\n\nClaim to verify: \"the date is $29\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p3": "0.444", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_1": "22", "answer_0": "40", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 06:30 to 17:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "10h30", "dur_3": "9h45", "dur_1": "10h45", "dur_0": "10h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1963. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "61", "age_2": "63", "age_0": "62", "age_3": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 365 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "140", "answer_0": "120", "answer_3": "127", "answer_1": "134"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $9k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1960. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "66", "age_1": "64", "age_3": "63", "age_0": "65"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Monday, April 13, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1986. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "40", "age_0": "39", "age_1": "38", "age_3": "41"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p1": "0.150", "p2": "0.050", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 338:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, February 06, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = $429; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1990. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "33", "age_0": "34", "age_3": "32", "age_2": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $655. A fee of $103 is charged, then a deposit of $74 arrives. A second fee of $51 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "554", "answer_3": "593", "answer_1": "546", "answer_0": "575"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = 2026-07-07; priority = high.\n\nClaim to verify: \"the date is 2026-07-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 284:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "51", "age_3": "50", "age_0": "52", "age_2": "53"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p1": "0.250", "p0": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, September 22, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1981. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "43", "age_3": "42", "age_2": "45", "age_0": "44"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1581 per year. It increases by 25% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1901.25", "answer_0": "1939.25", "answer_1": "1976.25", "answer_2": "1185.75"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = $324; priority = low.\n\nClaim to verify: \"the date is $324\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=410, unit_price=879, discount_pct=0, region=EU, sku_total=1289 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1289", "v2": "36039", "v0": "360390", "v3": "360800"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.615", "p0": "0.385", "p3": "0.333", "p1": "0.615"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.500", "p2": "0.333", "p1": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.286", "p0": "0.714", "p3": "0.667", "p1": "0.286"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed October 26, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 04", "date_2": "November 02", "date_1": "November 02", "date_3": "November 04"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, March 15, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = $335; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p1": "0.150", "p3": "0.050", "p0": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, August 01, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 72 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "212", "answer_1": "222", "answer_0": "224", "answer_2": "242"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p1": "0.250", "p3": "0.083", "p2": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "17", "answer_3": "21", "answer_0": "35", "answer_2": "14"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = $671; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1984. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "42", "age_0": "41", "age_1": "40", "age_3": "43"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.375", "p3": "0.125", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, January 27, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = $517; priority = high.\n\nClaim to verify: \"the amount is $517\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = $555; priority = low.\n\nClaim to verify: \"the amount is $555\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p0": "0.500", "p3": "0.125", "p1": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p1": "0.500", "p3": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p2": "0.571", "p3": "0.333", "p0": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Saturday, July 25, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 869:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "25", "answer_1": "35", "answer_0": "55", "answer_2": "30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1893 per year. It increases by 25% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2327.25", "answer_3": "2312.25", "answer_1": "2366.25", "answer_2": "1419.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p3": "0.050", "p0": "0.150", "p2": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=442, unit_price=675, discount_pct=0, region=EU, sku_total=1117 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "29835", "v3": "297908", "v1": "1117", "v0": "298350"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.700", "p0": "0.300", "p1": "0.700", "p3": "0.222"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=783, unit_price=792, discount_pct=0, region=EU, sku_total=1575 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1575", "v3": "619353", "v2": "62013", "v0": "620136"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 389 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "111", "answer_0": "105", "answer_3": "113", "answer_2": "125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=980, unit_price=868, discount_pct=0, region=EU, sku_total=1848 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "85064", "v3": "849660", "v0": "850640", "v1": "1848"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "20", "answer_2": "10", "answer_1": "12", "answer_3": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $178. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $824 per year. It increases by 15% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "947.6", "answer_2": "700.4", "answer_0": "900.6", "answer_3": "886.6"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_0": "24", "answer_3": "12", "answer_1": "14"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $625 per year. It increases by 5% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "593.75", "answer_1": "656.25", "answer_3": "622.25", "answer_0": "630.25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 239 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "54", "answer_1": "39", "answer_2": "72", "answer_3": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $204. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=807, unit_price=566, discount_pct=0, region=EU, sku_total=1373 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1373", "v3": "455955", "v2": "45676", "v0": "456762"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1974. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "49", "age_1": "50", "age_2": "52", "age_0": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.667", "p1": "0.667", "p3": "0.273"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.300", "p2": "0.700", "p3": "0.222", "p1": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed January 26, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 06", "date_0": "February 06", "date_2": "February 08", "date_1": "February 04"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 617:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 77 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_3": "58", "answer_2": "21", "answer_1": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=806, unit_price=166, discount_pct=0, region=EU, sku_total=972 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "132990", "v2": "13379", "v1": "972", "v0": "133796"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 142 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "9", "answer_2": "45", "answer_0": "28", "answer_3": "48"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p3": "0.250", "p1": "0.600", "p2": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "32", "answer_3": "28", "answer_0": "56", "answer_2": "28"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = 2026-07-24; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.250", "p0": "0.333", "p1": "0.417"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 16:00 to 22:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "6h00", "dur_3": "5h30", "dur_1": "6h30", "dur_0": "5h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = $31; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 7 grams of salt per serving and serves 12 people. A shopper bought a 84 gram box and a 840 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1815 per year. It increases by 20% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1452", "answer_1": "2178", "answer_3": "2173", "answer_0": "2144"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 307 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_3": "27", "answer_1": "31", "answer_0": "18"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, August 15, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed January 08, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 19", "date_1": "January 17", "date_2": "January 19", "date_0": "January 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 7 people. A shopper bought a 49 gram box and a 490 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.500", "p0": "0.375", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=678, unit_price=767, discount_pct=0, region=EU, sku_total=1445 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1445", "v3": "520704", "v2": "52002", "v0": "520026"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $10k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "66", "age_1": "63", "age_0": "64", "age_2": "65"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed December 19, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 25", "date_1": "December 23", "date_0": "December 24", "date_2": "December 22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.600", "p1": "0.600", "p0": "0.400", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $329. A fee of $127 is charged, then a deposit of $20 arrives. A second fee of $63 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "165", "answer_3": "169", "answer_0": "159", "answer_2": "184"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p2": "0.150", "p3": "0.050", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = 2026-10-09; priority = high.\n\nClaim to verify: \"the status is 2026-10-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 989:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, October 10, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $24k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 2 grams of salt per serving and serves 12 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = $738; priority = low.\n\nClaim to verify: \"the status is $738\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.500", "p3": "0.455", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 687:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $644 per year. It increases by 10% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "708.4", "answer_3": "663.4", "answer_2": "579.6", "answer_0": "697.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed August 11, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 18", "date_1": "August 16", "date_0": "August 18", "date_2": "August 20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $573. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.143", "p0": "0.250", "p1": "0.750", "p2": "0.750"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 6 grams of salt per serving and serves 7 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $902 per year. It increases by 20% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "721.6", "answer_3": "1016.4", "answer_1": "1082.4", "answer_0": "1051.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1279 per year. It increases by 5% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1301.95", "answer_1": "1342.95", "answer_3": "1335.95", "answer_2": "1215.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, July 06, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $595. A fee of $33 is charged, then a deposit of $90 arrives. A second fee of $16 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "613", "answer_0": "636", "answer_2": "646", "answer_3": "626"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1026 per year. It increases by 10% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1128.6", "answer_2": "923.4", "answer_0": "1114.6", "answer_3": "1106.6"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, January 28, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $12k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 608:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.533", "p0": "0.467", "p3": "0.429", "p1": "0.533"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.286", "p1": "0.714", "p2": "0.714", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $4k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = ops; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p0": "0.250", "p3": "0.125", "p1": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = 2026-02-25; priority = high.\n\nClaim to verify: \"the owner is 2026-02-25\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: date = $972; priority = low.\n\nClaim to verify: \"the date is $972\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: date = open; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1567 per year. It increases by 10% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1703.7", "answer_3": "1693.7", "answer_2": "1410.3", "answer_1": "1723.7"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "42", "answer_0": "84", "answer_1": "48", "answer_3": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Wednesday, December 02, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=912, unit_price=278, discount_pct=0, region=EU, sku_total=1190 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "252624", "v0": "253536", "v2": "25353", "v1": "1190"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = $222; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed April 06, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 15", "date_0": "April 13", "date_3": "April 13", "date_1": "April 11"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed February 12, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 21", "date_3": "February 23", "date_2": "February 27", "date_0": "February 25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Wednesday, May 06, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 06:45 to 21:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "14h30", "dur_3": "13h45", "dur_0": "14h00", "dur_1": "14h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = $598; priority = low.\n\nClaim to verify: \"the status is $598\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 8 grams of salt per serving and serves 11 people. A shopper bought a 88 gram box and a 880 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 5 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 9 grams of salt per serving and serves 11 people. A shopper bought a 99 gram box and a 990 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $822 per year. It increases by 25% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "616.5", "answer_0": "996.5", "answer_1": "1027.5", "answer_3": "1035.5"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 529:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=161, unit_price=575, discount_pct=0, region=EU, sku_total=736 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "92736", "v2": "9257", "v0": "92575", "v1": "736"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.167", "p1": "0.250", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1763 per year. It increases by 25% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2191.75", "answer_3": "2231.75", "answer_1": "2203.75", "answer_2": "1322.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 10 people. A shopper bought a 90 gram box and a 900 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.200", "p2": "0.727", "p1": "0.727", "p0": "0.273"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = $362; priority = low.\n\nClaim to verify: \"the amount is $362\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Thursday, May 28, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $17k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed September 13, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 19", "date_2": "September 23", "date_3": "September 21", "date_0": "September 21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1972. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "52", "age_0": "54", "age_2": "55", "age_1": "53"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.400", "p2": "0.400", "p0": "0.600", "p3": "0.556"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.667", "p3": "0.625", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $599. A fee of $128 is charged, then a deposit of $70 arrives. A second fee of $64 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "470", "answer_0": "477", "answer_2": "467", "answer_1": "456"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 303 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "160", "answer_1": "129", "answer_0": "140", "answer_3": "134"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_0": "24", "answer_1": "20", "answer_3": "6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $590. A fee of $199 is charged, then a deposit of $38 arrives. A second fee of $99 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "308", "answer_2": "345", "answer_0": "330", "answer_1": "324"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.571", "p0": "0.429", "p3": "0.385", "p1": "0.571"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "24", "answer_0": "42", "answer_3": "21", "answer_2": "21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $269. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=151, unit_price=809, discount_pct=0, region=EU, sku_total=960 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "122159", "v1": "960", "v2": "12215", "v3": "122310"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.500", "p3": "0.167", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 149 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "10", "answer_1": "7", "answer_0": "20", "answer_2": "29"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "65", "age_0": "64", "age_1": "63", "age_3": "66"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 347:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.125", "p2": "0.125", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 251 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "67", "answer_0": "54", "answer_1": "41", "answer_3": "64"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = 2026-12-14; priority = low.\n\nClaim to verify: \"the date is 2026-12-14\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1975. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "49", "age_1": "48", "age_2": "50", "age_3": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, October 20, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 118 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "182", "answer_2": "197", "answer_1": "199", "answer_0": "192"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = $492; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = closed; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1966. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "59", "age_0": "60", "age_3": "58", "age_2": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $314. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, December 07, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "49", "answer_2": "28", "answer_1": "31", "answer_3": "21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Tuesday, July 14, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1003 per year. It increases by 20% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1165.6", "answer_1": "1203.6", "answer_2": "802.4", "answer_0": "1155.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1332 per year. It increases by 20% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1550.4", "answer_2": "1065.6", "answer_0": "1580.4", "answer_1": "1598.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 8 grams of salt per serving and serves 5 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p2": "0.625", "p3": "0.286", "p0": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Wednesday, November 11, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = $267; priority = high.\n\nClaim to verify: \"the amount is $267\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=549, unit_price=316, discount_pct=0, region=EU, sku_total=865 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "173484", "v1": "865", "v2": "17348", "v3": "172935"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed May 02, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 08", "date_2": "May 05", "date_1": "May 06", "date_0": "May 07"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1975. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "50", "age_0": "49", "age_3": "47", "age_1": "48"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 829:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 311 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "44", "answer_2": "26", "answer_1": "32", "answer_0": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $311. A fee of $109 is charged, then a deposit of $74 arrives. A second fee of $54 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "215", "answer_1": "244", "answer_0": "222", "answer_3": "228"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = open; priority = high.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p2": "0.625", "p1": "0.625", "p3": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "54", "answer_3": "24", "answer_1": "34", "answer_2": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_1": "42", "answer_3": "36", "answer_2": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=187, unit_price=990, discount_pct=0, region=EU, sku_total=1177 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "185317", "v0": "185130", "v1": "1177", "v2": "18513"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 730:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 185 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "221", "answer_3": "205", "answer_1": "208", "answer_0": "210"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Friday, May 08, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = $103; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Sunday, April 19, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = $313; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_3": "9", "answer_1": "21", "answer_0": "27"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $9k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "31", "answer_2": "24", "answer_3": "56", "answer_0": "80"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 9 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 146 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "34", "answer_3": "73", "answer_0": "54", "answer_1": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $506. A fee of $158 is charged, then a deposit of $35 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "321", "answer_3": "283", "answer_2": "287", "answer_0": "304"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.050", "p0": "0.200", "p2": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, May 24, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=622, unit_price=986, discount_pct=0, region=EU, sku_total=1608 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "612670", "v1": "1608", "v0": "613292", "v2": "61329"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.250", "p0": "0.375", "p1": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1894 per year. It increases by 5% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1799.3", "answer_0": "1946.7", "answer_3": "1935.7", "answer_1": "1988.7"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 06:30 to 10:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "3h30", "dur_3": "2h30", "dur_2": "3h00", "dur_0": "2h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed August 14, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "August 24", "date_1": "August 20", "date_2": "August 22", "date_3": "August 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1444 per year. It increases by 5% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1516.2", "answer_2": "1371.8", "answer_0": "1468.2", "answer_3": "1480.2"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=140, unit_price=852, discount_pct=0, region=EU, sku_total=992 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "119280", "v2": "11928", "v1": "992", "v3": "119420"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.125", "p1": "0.500", "p0": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.200", "p0": "0.150", "p2": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.556", "p3": "0.375", "p2": "0.556", "p0": "0.444"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = 2026-01-10; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "27", "answer_0": "35", "answer_2": "25", "answer_3": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 633:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 15:15 to 15:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "0h00", "dur_1": "0h15", "dur_0": "-1h45", "dur_3": "-1h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed September 17, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "October 01", "date_3": "September 29", "date_2": "September 29", "date_1": "September 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.500", "p2": "0.167", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 177 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_1": "22", "answer_2": "26", "answer_3": "22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = 2026-08-21; priority = high.\n\nClaim to verify: \"the amount is 2026-08-21\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed November 04, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 10", "date_1": "November 08", "date_0": "November 10", "date_2": "November 08"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = 2026-03-26; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 275 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "152", "answer_1": "154", "answer_0": "150", "answer_2": "148"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $883. A fee of $166 is charged, then a deposit of $72 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "706", "answer_3": "724", "answer_1": "690", "answer_2": "696"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = pending; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.467", "p1": "0.467", "p0": "0.533"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 12:45 to 22:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "9h30", "dur_0": "9h00", "dur_3": "8h30", "dur_2": "9h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 312 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "100", "answer_2": "94", "answer_3": "96", "answer_1": "108"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1977. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "49", "age_3": "51", "age_2": "50", "age_1": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 395 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "125", "answer_1": "114", "answer_3": "107", "answer_2": "117"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.125", "p2": "0.375", "p1": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=340, unit_price=707, discount_pct=0, region=EU, sku_total=1047 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1047", "v2": "24038", "v3": "240720", "v0": "240380"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, September 24, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 243:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 5 grams of salt per serving and serves 3 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, April 09, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 992:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=486, unit_price=258, discount_pct=0, region=EU, sku_total=744 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "124902", "v2": "12538", "v1": "744", "v0": "125388"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1968. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "56", "age_2": "57", "age_3": "58", "age_1": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=940, unit_price=667, discount_pct=0, region=EU, sku_total=1607 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "626040", "v2": "62698", "v1": "1607", "v0": "626980"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "35", "answer_1": "23", "answer_2": "21", "answer_3": "14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 754:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 5 grams of salt per serving and serves 12 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $8k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $558. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 07:30 to 08:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h30", "dur_0": "0h15", "dur_1": "0h30", "dur_2": "0h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 969:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = $915; priority = low.\n\nClaim to verify: \"the amount is $915\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.200", "p2": "0.100", "p0": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A shift runs from 12:00 to 22:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "9h30", "dur_1": "10h00", "dur_2": "9h45", "dur_3": "9h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Friday, June 19, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=475, unit_price=152, discount_pct=0, region=EU, sku_total=627 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "627", "v3": "71725", "v2": "7220", "v0": "72200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.375", "p0": "0.250", "p2": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A recipe needs 2 grams of salt per serving and serves 10 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed December 01, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 05", "date_0": "December 07", "date_3": "December 07", "date_1": "December 05"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.462", "p0": "0.538", "p3": "0.500", "p2": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.100", "p1": "0.200", "p0": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $317 per year. It increases by 20% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "367.4", "answer_0": "336.4", "answer_2": "253.6", "answer_1": "380.4"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Contract signed September 14, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 19", "date_0": "September 21", "date_3": "September 21", "date_1": "September 19"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1990. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "34", "age_3": "37", "age_0": "35", "age_2": "36"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "22", "answer_2": "20", "answer_0": "28", "answer_3": "8"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p1": "0.150", "p3": "0.050", "p2": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.333", "p0": "0.250", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $741. A fee of $79 is charged, then a deposit of $20 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "626", "answer_2": "628", "answer_3": "656", "answer_0": "643"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 384:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.167", "p2": "0.167", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.167", "p3": "0.083", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p2": "0.500", "p3": "0.444"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p1": "0.250", "p3": "0.050", "p2": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 195 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "319", "answer_0": "336", "answer_1": "324", "answer_2": "350"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 272 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_1": "43", "answer_2": "40", "answer_3": "22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = $578; priority = high.\n\nClaim to verify: \"the owner is $578\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $492. A fee of $100 is charged, then a deposit of $87 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "419", "answer_0": "429", "answer_3": "418", "answer_1": "446"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1986. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "40", "age_1": "37", "age_0": "38", "age_2": "39"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, August 04, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 353 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_2": "86", "answer_3": "63", "answer_1": "65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 245 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "118", "answer_3": "102", "answer_0": "120", "answer_1": "115"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $39k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = $847; priority = high.\n\nClaim to verify: \"the status is $847\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "30", "answer_2": "12", "answer_1": "15", "answer_3": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $447 per year. It increases by 20% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "357.6", "answer_1": "536.4", "answer_3": "532.4", "answer_0": "518.4"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.583", "p0": "0.417", "p3": "0.364", "p2": "0.583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=542, unit_price=564, discount_pct=0, region=EU, sku_total=1106 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "30568", "v3": "305146", "v0": "305688", "v1": "1106"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p2": "0.100", "p0": "0.150", "p3": "0.050"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=446, unit_price=680, discount_pct=0, region=EU, sku_total=1126 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "30328", "v3": "303726", "v0": "303280", "v1": "1126"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p3": "0.333", "p1": "0.600", "p2": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p3": "0.167", "p1": "0.667", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed July 18, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 27", "date_0": "July 30", "date_2": "July 28", "date_3": "July 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.625", "p2": "0.375", "p3": "0.571", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "14", "answer_3": "15", "answer_0": "24", "answer_2": "9"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $743. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.700", "p3": "0.222", "p0": "0.300", "p2": "0.700"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.714", "p0": "0.286", "p1": "0.714"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 995:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p0": "0.500", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 5 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=380, unit_price=741, discount_pct=0, region=EU, sku_total=1121 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "28158", "v0": "281580", "v1": "1121", "v3": "281960"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "96", "answer_2": "40", "answer_3": "56", "answer_1": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1963 per year. It increases by 25% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2453.75", "answer_0": "2425.75", "answer_3": "2398.75", "answer_2": "1472.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, October 18, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 7 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 08:30 to 18:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "8h30", "dur_1": "9h30", "dur_2": "9h15", "dur_0": "9h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=715, unit_price=373, discount_pct=0, region=EU, sku_total=1088 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "267410", "v2": "26669", "v1": "1088", "v0": "266695"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = $339; priority = low.\n\nClaim to verify: \"the date is $339\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "11", "answer_0": "15", "answer_3": "6", "answer_2": "9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=738, unit_price=482, discount_pct=0, region=EU, sku_total=1220 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "35571", "v1": "1220", "v0": "355716", "v3": "356454"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p2": "0.100", "p0": "0.150", "p3": "0.050"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "19", "answer_0": "40", "answer_2": "12", "answer_3": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.150", "p1": "0.250", "p0": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_0": "27", "answer_3": "15", "answer_1": "17"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Thursday, August 06, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 287 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_2": "90", "answer_3": "82", "answer_1": "88"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1963. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "65", "age_0": "63", "age_1": "62", "age_2": "64"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 8 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = 2026-11-24; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $464. A fee of $148 is charged, then a deposit of $61 arrives. A second fee of $74 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "303", "answer_2": "330", "answer_1": "289", "answer_3": "284"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.615", "p3": "0.333", "p1": "0.615", "p0": "0.385"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 320 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "22", "answer_1": "13", "answer_0": "32", "answer_3": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $573. A fee of $82 is charged, then a deposit of $80 arrives. A second fee of $41 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "501", "answer_3": "556", "answer_2": "536", "answer_0": "530"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 385 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "87", "answer_1": "96", "answer_3": "78", "answer_0": "90"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.167", "p0": "0.333", "p3": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 949:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 2 grams of salt per serving and serves 9 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "21", "answer_3": "49", "answer_0": "70", "answer_1": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1962. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "64", "age_3": "65", "age_1": "62", "age_0": "63"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_2": "18", "answer_1": "20", "answer_0": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = 2026-03-21; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=889, unit_price=612, discount_pct=0, region=EU, sku_total=1501 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "543179", "v1": "1501", "v2": "54406", "v0": "544068"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "22", "answer_2": "16", "answer_3": "48", "answer_0": "64"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = $613; priority = high.\n\nClaim to verify: \"the amount is $613\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.429", "p0": "0.571", "p3": "0.500", "p2": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = $74; priority = low.\n\nClaim to verify: \"the owner is $74\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = 2026-12-06; priority = low.\n\nClaim to verify: \"the date is 2026-12-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=841, unit_price=526, discount_pct=0, region=EU, sku_total=1367 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "442366", "v2": "44236", "v3": "443207", "v1": "1367"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $288. A fee of $59 is charged, then a deposit of $90 arrives. A second fee of $29 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "298", "answer_1": "301", "answer_0": "290", "answer_3": "302"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.050", "p0": "0.200", "p2": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An account starts at $509. A fee of $84 is charged, then a deposit of $36 arrives. A second fee of $42 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "419", "answer_2": "435", "answer_1": "409", "answer_3": "389"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.222", "p2": "0.222", "p0": "0.778", "p3": "0.750"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_0": "36", "answer_3": "18", "answer_1": "21"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=482, unit_price=585, discount_pct=0, region=EU, sku_total=1067 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "281970", "v2": "28197", "v1": "1067", "v3": "282452"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "30", "answer_1": "14", "answer_3": "20", "answer_2": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=429, unit_price=371, discount_pct=0, region=EU, sku_total=800 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "15915", "v3": "158730", "v1": "800", "v0": "159159"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "27", "answer_1": "19", "answer_3": "12", "answer_2": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=676, unit_price=767, discount_pct=0, region=EU, sku_total=1443 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "517816", "v1": "1443", "v2": "51849", "v0": "518492"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p2": "0.250", "p3": "0.083", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=678, unit_price=711, discount_pct=0, region=EU, sku_total=1389 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "48205", "v3": "481380", "v1": "1389", "v0": "482058"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 857:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = $217; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $815. A fee of $92 is charged, then a deposit of $89 arrives. A second fee of $46 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "766", "answer_1": "774", "answer_3": "779", "answer_2": "774"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 95 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "79", "answer_1": "55", "answer_0": "72", "answer_3": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = 2026-04-18; priority = high.\n\nClaim to verify: \"the owner is 2026-04-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $686. A fee of $95 is charged, then a deposit of $53 arrives. A second fee of $47 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "574", "answer_2": "568", "answer_1": "624", "answer_0": "597"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 795:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $36k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = 2026-03-13; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 160 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "146", "answer_1": "118", "answer_2": "123", "answer_0": "126"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 6 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $736. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.833", "p2": "0.500", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=247, unit_price=445, discount_pct=0, region=EU, sku_total=692 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "692", "v0": "109915", "v2": "10991", "v3": "110162"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 351 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "247", "answer_2": "221", "answer_3": "246", "answer_0": "240"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $19k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.083", "p0": "0.250", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = $763; priority = low.\n\nClaim to verify: \"the status is $763\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1990. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "36", "age_1": "34", "age_3": "33", "age_0": "35"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.050", "p0": "0.100", "p1": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=980, unit_price=788, discount_pct=0, region=EU, sku_total=1768 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "772240", "v2": "77224", "v3": "771260", "v1": "1768"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $455 per year. It increases by 10% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "502.5", "answer_0": "471.5", "answer_1": "500.5", "answer_2": "409.5"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = 2026-05-15; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 908:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 08:30 to 21:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "12h45", "dur_0": "12h15", "dur_3": "11h45", "dur_2": "12h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 287 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_2": "64", "answer_1": "77", "answer_3": "89"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.167", "p3": "0.083", "p2": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = 2026-08-19; priority = high.\n\nClaim to verify: \"the amount is 2026-08-19\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Sunday, December 27, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p1": "0.200", "p0": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 09:15 to 22:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "13h30", "dur_0": "13h15", "dur_3": "12h30", "dur_1": "13h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.286", "p2": "0.286", "p3": "0.667", "p0": "0.714"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $229. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p3": "0.167", "p0": "0.500", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 769:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, November 16, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed September 11, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 22", "date_1": "September 20", "date_0": "September 24", "date_2": "September 22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 9 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 581:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 4 grams of salt per serving and serves 7 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 739:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1769 per year. It increases by 25% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2211.25", "answer_2": "1326.75", "answer_3": "2203.25", "answer_0": "2163.25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=374, unit_price=769, discount_pct=0, region=EU, sku_total=1143 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "287606", "v1": "1143", "v2": "28760", "v3": "287232"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = $609; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 237 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "63", "answer_3": "73", "answer_0": "80", "answer_1": "95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 174 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "63", "answer_0": "56", "answer_1": "66", "answer_2": "75"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $739. A fee of $125 is charged, then a deposit of $30 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "600", "answer_1": "604", "answer_2": "552", "answer_0": "582"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = $415; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=863, unit_price=913, discount_pct=0, region=EU, sku_total=1776 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "788782", "v0": "787919", "v1": "1776", "v2": "78791"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p0": "0.625", "p3": "0.571", "p1": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1972. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "53", "age_3": "52", "age_0": "54", "age_2": "55"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = $644; priority = high.\n\nClaim to verify: \"the date is $644\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = $260; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed May 13, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 24", "date_2": "May 24", "date_1": "May 22", "date_0": "May 26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = 2026-11-10; priority = high.\n\nClaim to verify: \"the owner is 2026-11-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, October 09, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 345 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "118", "answer_0": "100", "answer_3": "109", "answer_1": "80"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1982. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "43", "age_2": "44", "age_3": "41", "age_1": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.083", "p0": "0.333", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 382 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "168", "answer_2": "187", "answer_3": "170", "answer_1": "180"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 4 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=203, unit_price=784, discount_pct=0, region=EU, sku_total=987 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "987", "v2": "15915", "v3": "159355", "v0": "159152"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p2": "0.500", "p3": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = 2026-09-18; priority = high.\n\nClaim to verify: \"the amount is 2026-09-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1960. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "66", "age_3": "63", "age_1": "64", "age_0": "65"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.444", "p0": "0.556", "p3": "0.500", "p2": "0.444"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = $19; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.545", "p0": "0.455", "p3": "0.400", "p1": "0.545"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = 2026-03-01; priority = low.\n\nClaim to verify: \"the amount is 2026-03-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1426 per year. It increases by 10% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1547.6", "answer_0": "1555.6", "answer_1": "1568.6", "answer_2": "1283.4"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 8 grams of salt per serving and serves 10 people. A shopper bought a 80 gram box and a 800 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = 2026-05-06; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 267 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_3": "107", "answer_1": "102", "answer_2": "131"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.375", "p3": "0.125", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "28", "answer_2": "16", "answer_3": "12", "answer_1": "19"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed May 25, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 28", "date_2": "May 30", "date_3": "May 30", "date_0": "May 28"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p2": "0.100", "p3": "0.050", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $507. A fee of $163 is charged, then a deposit of $42 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "316", "answer_3": "280", "answer_2": "307", "answer_0": "305"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.200", "p0": "0.273", "p1": "0.727", "p2": "0.727"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p3": "0.571", "p1": "0.375", "p0": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 450:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $34k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "12", "answer_2": "6", "answer_0": "24", "answer_3": "18"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 58 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "50", "answer_1": "57", "answer_3": "44", "answer_2": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed March 21, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 25", "date_2": "March 24", "date_3": "March 27", "date_0": "March 26"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A shift runs from 15:45 to 20:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "4h30", "dur_3": "3h30", "dur_1": "4h30", "dur_0": "4h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = 2026-04-27; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "36", "answer_3": "12", "answer_0": "48", "answer_1": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_0": "24", "answer_2": "6", "answer_1": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 196 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "216", "answer_1": "201", "answer_2": "236", "answer_3": "231"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1973. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "54", "age_0": "53", "age_1": "52", "age_3": "55"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 623:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.100", "p1": "0.200", "p0": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Thursday, October 22, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.273", "p1": "0.667", "p0": "0.333", "p2": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=802, unit_price=444, discount_pct=0, region=EU, sku_total=1246 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "35608", "v1": "1246", "v3": "356890", "v0": "356088"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1977. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "45", "age_2": "48", "age_1": "46", "age_0": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.200", "p0": "0.150", "p2": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $656. A fee of $132 is charged, then a deposit of $56 arrives. A second fee of $66 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "514", "answer_2": "525", "answer_1": "529", "answer_3": "501"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.727", "p3": "0.200", "p0": "0.273", "p2": "0.727"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $303 per year. It increases by 10% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "272.7", "answer_1": "333.3", "answer_0": "320.3", "answer_3": "337.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.429", "p0": "0.500", "p2": "0.500", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = 2026-12-01; priority = low.\n\nClaim to verify: \"the owner is 2026-12-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed September 26, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 03", "date_0": "October 06", "date_2": "October 04", "date_3": "October 05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $38k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=284, unit_price=451, discount_pct=0, region=EU, sku_total=735 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "735", "v3": "127800", "v0": "128084", "v2": "12808"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Form fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.462", "p1": "0.462", "p0": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = 2026-08-23; priority = low.\n\nClaim to verify: \"the amount is 2026-08-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 11:15 to 11:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "0h15", "dur_3": "-1h15", "dur_2": "0h30", "dur_0": "0h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1181 per year. It increases by 5% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1220.05", "answer_2": "1121.95", "answer_3": "1239.05", "answer_1": "1240.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1601 per year. It increases by 25% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1200.75", "answer_3": "1960.25", "answer_1": "2001.25", "answer_0": "1976.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.500", "p1": "0.667", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p2": "0.050", "p0": "0.100", "p3": "0.050"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $884. A fee of $64 is charged, then a deposit of $90 arrives. A second fee of $32 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "878", "answer_2": "897", "answer_1": "887", "answer_3": "868"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1334 per year. It increases by 15% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1534.1", "answer_0": "1488.1", "answer_2": "1133.9", "answer_3": "1477.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Sunday, September 13, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 769:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = $351; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 902:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 12:00 to 19:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "7h00", "dur_0": "6h45", "dur_1": "7h30", "dur_3": "6h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $8k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.125", "p3": "0.125", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed December 10, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 16", "date_1": "December 14", "date_0": "December 16", "date_2": "December 14"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $597. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.600", "p1": "0.364", "p0": "0.636", "p2": "0.364"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "20", "answer_0": "35", "answer_3": "15", "answer_1": "23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Wednesday, November 11, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed January 24, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 29", "date_2": "February 01", "date_3": "January 31", "date_0": "January 30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.167", "p3": "0.083", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.545", "p0": "0.455", "p3": "0.400", "p1": "0.545"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=550, unit_price=406, discount_pct=0, region=EU, sku_total=956 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "956", "v2": "22330", "v0": "223300", "v3": "223850"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 263:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $264. A fee of $133 is charged, then a deposit of $50 arrives. A second fee of $66 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "115", "answer_3": "129", "answer_2": "138", "answer_1": "118"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $334. A fee of $101 is charged, then a deposit of $80 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "275", "answer_3": "249", "answer_2": "285", "answer_0": "263"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 4 grams of salt per serving and serves 6 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "52", "answer_2": "24", "answer_3": "28", "answer_1": "31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=537, unit_price=220, discount_pct=0, region=EU, sku_total=757 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "11814", "v1": "757", "v0": "118140", "v3": "118677"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 786:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 3 grams of salt per serving and serves 7 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $437. A fee of $121 is charged, then a deposit of $37 arrives. A second fee of $60 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "275", "answer_2": "302", "answer_1": "316", "answer_0": "293"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 284 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_2": "17", "answer_1": "26", "answer_0": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $612. A fee of $129 is charged, then a deposit of $35 arrives. A second fee of $64 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "451", "answer_2": "433", "answer_0": "454", "answer_3": "446"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 247 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "172", "answer_0": "180", "answer_2": "197", "answer_3": "194"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = 2026-11-08; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $466. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $838. A fee of $46 is charged, then a deposit of $47 arrives. A second fee of $23 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "816", "answer_2": "797", "answer_3": "809", "answer_1": "802"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed March 07, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 18", "date_3": "March 17", "date_2": "March 16", "date_1": "March 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 419:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "19", "answer_3": "21", "answer_0": "33", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 449:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed May 25, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 30", "date_1": "May 28", "date_3": "May 30", "date_0": "May 28"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = open; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 9 grams of salt per serving and serves 3 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Sunday, March 08, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $34k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = 2026-11-04; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = pending; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = 2026-05-05; priority = high.\n\nClaim to verify: \"the owner is 2026-05-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed July 20, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "August 02", "date_1": "July 29", "date_0": "July 31", "date_3": "July 31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed September 13, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 23", "date_1": "September 23", "date_0": "September 25", "date_3": "September 25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = 2026-09-14; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 438:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = 2026-01-07; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=240, unit_price=535, discount_pct=0, region=EU, sku_total=775 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "12840", "v1": "775", "v3": "128160", "v0": "128400"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = ops; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p0": "0.500", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p3": "0.167", "p2": "0.500", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 385 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "144", "answer_2": "134", "answer_3": "136", "answer_1": "133"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 7 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, May 05, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $424. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $7k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1446 per year. It increases by 15% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1662.9", "answer_3": "1623.9", "answer_0": "1635.9", "answer_2": "1229.1"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1958. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "67", "age_2": "69", "age_3": "70", "age_0": "68"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 126 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "117", "answer_0": "120", "answer_2": "107", "answer_3": "113"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 06:15 to 17:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "10h45", "dur_1": "11h15", "dur_2": "11h15", "dur_3": "10h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $860 per year. It increases by 20% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1032", "answer_0": "990", "answer_3": "1000", "answer_2": "688"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, September 07, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1971. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "53", "age_1": "54", "age_0": "55", "age_2": "56"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $18k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = $157; priority = high.\n\nClaim to verify: \"the date is $157\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = ops; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "19", "answer_2": "12", "answer_0": "54", "answer_3": "42"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $492. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $697 per year. It increases by 10% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "627.3", "answer_3": "727.7", "answer_0": "719.7", "answer_1": "766.7"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $613. A fee of $160 is charged, then a deposit of $54 arrives. A second fee of $80 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "427", "answer_2": "452", "answer_3": "400", "answer_1": "422"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p3": "0.167", "p1": "0.667", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $5k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=425, unit_price=659, discount_pct=0, region=EU, sku_total=1084 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "280500", "v1": "1084", "v2": "28007", "v0": "280075"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 19:15 to 22:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "3h30", "dur_3": "2h30", "dur_0": "3h00", "dur_2": "3h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $754 per year. It increases by 15% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "828.1", "answer_2": "640.9", "answer_1": "867.1", "answer_0": "850.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "64", "answer_1": "22", "answer_2": "16", "answer_3": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1370 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1406.5", "answer_1": "1438.5", "answer_3": "1434.5", "answer_2": "1301.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=369, unit_price=836, discount_pct=0, region=EU, sku_total=1205 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "308853", "v1": "1205", "v2": "30848", "v0": "308484"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $671. A fee of $136 is charged, then a deposit of $54 arrives. A second fee of $68 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "551", "answer_0": "521", "answer_3": "548", "answer_2": "547"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed July 05, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 16", "date_0": "July 14", "date_1": "July 12", "date_3": "July 14"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 744:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 137 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "96", "answer_1": "77", "answer_0": "90", "answer_2": "82"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p3": "0.050", "p2": "0.050", "p0": "0.100"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 11:45 to 17:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "4h45", "dur_1": "5h45", "dur_2": "5h45", "dur_0": "5h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 354 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "16", "answer_2": "43", "answer_3": "22", "answer_0": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 371:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1988. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "38", "age_3": "39", "age_1": "36", "age_0": "37"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed May 28, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 02", "date_1": "May 31", "date_2": "June 04", "date_0": "June 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Wednesday, March 04, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 772:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1967. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "57", "age_2": "58", "age_3": "59", "age_1": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.545", "p0": "0.455", "p3": "0.400", "p1": "0.545"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $805. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p1": "0.500", "p3": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "20", "answer_2": "14", "answer_3": "42", "answer_0": "56"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $26k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p0": "0.583", "p2": "0.417", "p3": "0.545"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=660, unit_price=461, discount_pct=0, region=EU, sku_total=1121 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "303600", "v1": "1121", "v2": "30426", "v0": "304260"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = $248; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, October 02, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "30", "answer_2": "10", "answer_1": "16", "answer_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 5 grams of salt per serving and serves 5 people. A shopper bought a 25 gram box and a 250 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.333", "p3": "0.636", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $587. A fee of $168 is charged, then a deposit of $26 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "371", "answer_0": "361", "answer_3": "373", "answer_2": "385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 687:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=195, unit_price=400, discount_pct=0, region=EU, sku_total=595 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "78000", "v1": "595", "v3": "78195", "v2": "7800"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p1": "0.250", "p0": "0.200", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $889. A fee of $143 is charged, then a deposit of $72 arrives. A second fee of $71 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "726", "answer_3": "730", "answer_1": "731", "answer_0": "747"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 08:45 to 18:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "9h00", "dur_1": "9h45", "dur_3": "8h45", "dur_2": "9h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1959. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "66", "age_2": "68", "age_0": "67", "age_3": "69"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1985. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "41", "age_3": "42", "age_0": "40", "age_1": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 283 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "160", "answer_3": "173", "answer_1": "178", "answer_2": "174"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 6 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = 2026-07-06; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 10:30 to 16:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "5h45", "dur_2": "6h00", "dur_3": "5h00", "dur_1": "6h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.222", "p2": "0.778", "p1": "0.778"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, May 02, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.714", "p0": "0.286", "p2": "0.714"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1962. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "63", "age_3": "61", "age_1": "62", "age_2": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1609 per year. It increases by 15% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1811.35", "answer_3": "1772.35", "answer_2": "1367.65", "answer_1": "1850.35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = 2026-05-20; priority = low.\n\nClaim to verify: \"the status is 2026-05-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 980:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 413:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, August 04, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p3": "0.050", "p0": "0.150", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 293 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "122", "answer_1": "111", "answer_0": "126", "answer_2": "107"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, January 16, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 16:30 to 20:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "3h45", "dur_3": "3h00", "dur_1": "4h00", "dur_2": "4h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=889, unit_price=334, discount_pct=0, region=EU, sku_total=1223 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "29692", "v0": "296926", "v1": "1223", "v3": "297815"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = 2026-07-23; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = $612; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = 2026-01-04; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "16", "answer_0": "56", "answer_2": "40", "answer_1": "42"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "26", "answer_2": "21", "answer_0": "56", "answer_3": "35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1990. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "34", "age_2": "35", "age_1": "33", "age_3": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $375. A fee of $103 is charged, then a deposit of $29 arrives. A second fee of $51 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "250", "answer_2": "248", "answer_3": "271", "answer_1": "277"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $219. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 07:45 to 10:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "2h30", "dur_1": "2h45", "dur_3": "1h45", "dur_0": "2h00"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.167", "p2": "0.500", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 101 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "116", "answer_3": "100", "answer_0": "96", "answer_2": "103"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 280 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "171", "answer_3": "167", "answer_1": "151", "answer_0": "162"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p2": "0.600", "p1": "0.600", "p0": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=750, unit_price=253, discount_pct=0, region=EU, sku_total=1003 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "18975", "v3": "189000", "v1": "1003", "v0": "189750"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed January 03, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 13", "date_2": "January 16", "date_0": "January 14", "date_1": "January 11"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.286", "p2": "0.625", "p0": "0.375", "p1": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = $441; priority = high.\n\nClaim to verify: \"the status is $441\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.455", "p2": "0.455", "p3": "0.500", "p0": "0.545"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 9 grams of salt per serving and serves 3 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = $15; priority = low.\n\nClaim to verify: \"the status is $15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.286", "p0": "0.714", "p3": "0.667", "p1": "0.286"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, November 27, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1980. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "45", "age_0": "46", "age_3": "48", "age_2": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 08:00 to 21:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "13h45", "dur_0": "13h00", "dur_3": "12h45", "dur_2": "13h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 270:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = 2026-10-15; priority = high.\n\nClaim to verify: \"the amount is 2026-10-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 292 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "71", "answer_1": "62", "answer_2": "64", "answer_0": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1400 per year. It increases by 15% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1597", "answer_3": "1589", "answer_1": "1610", "answer_2": "1190"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 328 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "52", "answer_3": "22", "answer_0": "36", "answer_2": "42"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $303. A fee of $41 is charged, then a deposit of $40 arrives. A second fee of $20 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "262", "answer_3": "259", "answer_1": "269", "answer_0": "282"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = 2026-07-17; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = $221; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 384 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "106", "answer_2": "113", "answer_1": "129", "answer_0": "120"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $174. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1281 per year. It increases by 20% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1024.8", "answer_0": "1519.2", "answer_3": "1510.2", "answer_1": "1537.2"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.385", "p0": "0.615", "p2": "0.385", "p3": "0.583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 661:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1362 per year. It increases by 20% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1623.4", "answer_2": "1089.6", "answer_3": "1645.4", "answer_1": "1634.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.083", "p2": "0.250", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $400. A fee of $92 is charged, then a deposit of $40 arrives. A second fee of $46 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "302", "answer_3": "299", "answer_2": "274", "answer_1": "293"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $505 per year. It increases by 10% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "497.5", "answer_1": "555.5", "answer_0": "535.5", "answer_2": "454.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.375", "p3": "0.125", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $680. A fee of $92 is charged, then a deposit of $72 arrives. A second fee of $46 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "603", "answer_0": "614", "answer_1": "622", "answer_3": "627"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 5 people. A shopper bought a 25 gram box and a 250 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 281:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 630:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $561. A fee of $183 is charged, then a deposit of $49 arrives. A second fee of $91 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "329", "answer_1": "327", "answer_0": "336", "answer_3": "323"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = $220; priority = low.\n\nClaim to verify: \"the status is $220\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed November 19, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 01", "date_2": "December 01", "date_1": "November 29", "date_0": "December 03"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed September 27, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 30", "date_2": "September 28", "date_3": "October 02", "date_1": "September 30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1459 per year. It increases by 20% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1734.8", "answer_3": "1742.8", "answer_1": "1750.8", "answer_2": "1167.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.500", "p0": "0.375", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1979. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "46", "age_0": "45", "age_3": "43", "age_1": "44"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = 2026-03-07; priority = high.\n\nClaim to verify: \"the owner is 2026-03-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.250", "p1": "0.417", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=331, unit_price=700, discount_pct=0, region=EU, sku_total=1031 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "232031", "v1": "1031", "v2": "23170", "v0": "231700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.333", "p3": "0.083", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p3": "0.050", "p2": "0.150", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = $754; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.333", "p0": "0.500", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed July 17, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 20", "date_3": "July 22", "date_0": "July 22", "date_2": "July 24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Thursday, December 10, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Monday, March 16, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p2": "0.375", "p3": "0.125", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1978. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "46", "age_3": "48", "age_1": "45", "age_2": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 798:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=420, unit_price=878, discount_pct=0, region=EU, sku_total=1298 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "36876", "v0": "368760", "v1": "1298", "v3": "369180"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1985. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "40", "age_3": "39", "age_0": "41", "age_2": "42"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1090 per year. It increases by 5% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1138.5", "answer_2": "1035.5", "answer_1": "1144.5", "answer_0": "1125.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "20", "answer_2": "10", "answer_1": "12", "answer_3": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $31k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "33", "answer_3": "35", "answer_2": "28", "answer_0": "63"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 17:30 to 19:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h00", "dur_0": "1h45", "dur_3": "1h00", "dur_2": "2h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=860, unit_price=908, discount_pct=0, region=EU, sku_total=1768 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "780880", "v3": "781740", "v2": "78088", "v1": "1768"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 07:00 to 14:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "6h30", "dur_2": "7h00", "dur_1": "7h15", "dur_3": "6h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p3": "0.050", "p2": "0.100", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.500", "p1": "0.833", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = 2026-02-14; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: amount = 2026-10-16; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 241 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "55", "answer_3": "63", "answer_1": "43", "answer_0": "48"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = closed; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 264 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "69", "answer_0": "60", "answer_1": "50", "answer_2": "54"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = 2026-05-23; priority = high.\n\nClaim to verify: \"the status is 2026-05-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $343. A fee of $84 is charged, then a deposit of $83 arrives. A second fee of $42 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "300", "answer_2": "287", "answer_1": "284", "answer_3": "320"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1972. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "52", "age_2": "53", "age_1": "51", "age_3": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $18k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1753 per year. It increases by 10% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1577.7", "answer_1": "1928.3", "answer_0": "1913.3", "answer_3": "1896.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=536, unit_price=327, discount_pct=0, region=EU, sku_total=863 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "17527", "v3": "175808", "v0": "175272", "v1": "863"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $793 per year. It increases by 15% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "868.95", "answer_1": "911.95", "answer_2": "674.05", "answer_3": "840.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 341:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = $676; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.500", "p2": "0.167", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, September 16, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed February 28, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 11", "date_3": "March 08", "date_0": "March 09", "date_1": "March 06"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_0": "77", "answer_1": "47", "answer_2": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_2": "12", "answer_3": "24", "answer_1": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = 2026-08-10; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $198. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1966. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "61", "age_1": "59", "age_3": "58", "age_0": "60"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.667", "p1": "0.667", "p3": "0.273"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $412. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "8", "answer_0": "32", "answer_2": "24", "answer_1": "26"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=621, unit_price=943, discount_pct=0, region=EU, sku_total=1564 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "585603", "v1": "1564", "v2": "58560", "v3": "584982"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1847 per year. It increases by 20% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2194.4", "answer_0": "2183.4", "answer_2": "1477.6", "answer_1": "2216.4"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=892, unit_price=176, discount_pct=0, region=EU, sku_total=1068 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "157884", "v0": "156992", "v1": "1068", "v2": "15699"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "42", "answer_2": "12", "answer_3": "30", "answer_1": "17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = $546; priority = high.\n\nClaim to verify: \"the amount is $546\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 301 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "92", "answer_0": "75", "answer_3": "77", "answer_2": "82"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1970. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "56", "age_3": "53", "age_0": "55", "age_1": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Sunday, November 22, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 464:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1989. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "38", "age_0": "37", "age_3": "35", "age_1": "36"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $854. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, April 10, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 532:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: date = ops; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = $143; priority = low.\n\nClaim to verify: \"the status is $143\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.727", "p0": "0.273", "p1": "0.727", "p3": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1606 per year. It increases by 15% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1811.9", "answer_0": "1817.9", "answer_2": "1365.1", "answer_1": "1846.9"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $604. A fee of $166 is charged, then a deposit of $71 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "451", "answer_2": "414", "answer_3": "401", "answer_0": "426"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 8 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p3": "0.050", "p0": "0.150", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed April 11, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 21", "date_0": "April 22", "date_1": "April 19", "date_2": "April 24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 790:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, August 07, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Wednesday, August 05, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = 2026-08-02; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1967. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "59", "age_3": "57", "age_2": "60", "age_1": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 621:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p1": "0.250", "p3": "0.050", "p0": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = 2026-12-25; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.500", "p2": "0.500", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.625", "p2": "0.333", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = 2026-12-27; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $27k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=302, unit_price=162, discount_pct=0, region=EU, sku_total=464 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "48622", "v0": "48924", "v2": "4892", "v1": "464"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A recipe needs 3 grams of salt per serving and serves 8 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=243, unit_price=350, discount_pct=0, region=EU, sku_total=593 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "84807", "v2": "8505", "v1": "593", "v0": "85050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 8 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1039 per year. It increases by 20% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1246.8", "answer_0": "1220.8", "answer_2": "831.2", "answer_3": "1243.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 210 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "192", "answer_3": "196", "answer_1": "204", "answer_2": "206"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = 2026-12-20; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=289, unit_price=304, discount_pct=0, region=EU, sku_total=593 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "87856", "v1": "593", "v3": "87567", "v2": "8785"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 894:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1962. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "61", "age_0": "62", "age_2": "63", "age_3": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.462", "p0": "0.538", "p3": "0.500", "p2": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1979. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "46", "age_3": "45", "age_2": "48", "age_0": "47"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p2": "0.571", "p3": "0.333", "p0": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.500", "p3": "0.167", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1977. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "50", "age_0": "49", "age_3": "51", "age_1": "48"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 306:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = $301; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = $307; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p3": "0.050", "p1": "0.200", "p0": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p1": "0.250", "p0": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = ops; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=660, unit_price=452, discount_pct=0, region=EU, sku_total=1112 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1112", "v2": "29832", "v0": "298320", "v3": "298980"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = $216; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.600", "p1": "0.333", "p0": "0.667", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p1": "0.600", "p2": "0.600", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_3": "24", "answer_1": "19", "answer_0": "40"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $148. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.167", "p0": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=440, unit_price=213, discount_pct=0, region=EU, sku_total=653 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "94160", "v1": "653", "v2": "9372", "v0": "93720"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 743:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $224. A fee of $56 is charged, then a deposit of $68 arrives. A second fee of $28 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "208", "answer_2": "228", "answer_1": "238", "answer_3": "188"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.429", "p0": "0.571", "p2": "0.429", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed September 23, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 29", "date_2": "September 27", "date_3": "September 29", "date_1": "September 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = 2026-12-25; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 447:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $644 per year. It increases by 10% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "680.4", "answer_2": "579.6", "answer_0": "685.4", "answer_1": "708.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $580. A fee of $51 is charged, then a deposit of $23 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "527", "answer_3": "501", "answer_1": "534", "answer_2": "553"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 11 people. A shopper bought a 88 gram box and a 880 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $36k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed October 28, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 02", "date_0": "November 04", "date_3": "November 04", "date_2": "November 06"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.667", "p3": "0.636", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1728 per year. It increases by 10% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1555.2", "answer_1": "1900.8", "answer_3": "1841.8", "answer_0": "1856.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 489:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1978. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "48", "age_3": "50", "age_2": "49", "age_1": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.556", "p2": "0.556", "p3": "0.375", "p0": "0.444"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.385", "p2": "0.615", "p1": "0.615", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.333", "p1": "0.667", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, January 05, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.250", "p3": "0.714", "p0": "0.750"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1989. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "35", "age_2": "37", "age_0": "36", "age_3": "34"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $667 per year. It increases by 25% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "500.25", "answer_3": "799.75", "answer_1": "833.75", "answer_0": "791.75"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: amount = 2026-05-26; priority = high.\n\nClaim to verify: \"the amount is 2026-05-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "18", "answer_2": "16", "answer_3": "16", "answer_0": "32"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = 2026-04-21; priority = low.\n\nClaim to verify: \"the owner is 2026-04-21\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p1": "0.250", "p3": "0.083", "p0": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = $575; priority = low.\n\nClaim to verify: \"the status is $575\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = 2026-05-21; priority = low.\n\nClaim to verify: \"the owner is 2026-05-21\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $172. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 477:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, April 08, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 246 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "160", "answer_1": "148", "answer_3": "153", "answer_2": "178"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = $741; priority = low.\n\nClaim to verify: \"the date is $741\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1308 per year. It increases by 20% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1514.6", "answer_1": "1569.6", "answer_0": "1553.6", "answer_2": "1046.4"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = 2026-04-03; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: date = ops; priority = low.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Monday, January 05, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $5k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 2 grams of salt per serving and serves 12 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=150, unit_price=986, discount_pct=0, region=EU, sku_total=1136 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "147900", "v3": "147750", "v1": "1136", "v2": "14790"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = closed; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 19:45 to 20:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "1h00", "dur_2": "0h45", "dur_3": "0h00", "dur_0": "0h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 243 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "127", "answer_1": "127", "answer_0": "144", "answer_2": "161"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 279 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "123", "answer_1": "103", "answer_3": "121", "answer_0": "105"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $508. A fee of $189 is charged, then a deposit of $90 arrives. A second fee of $94 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "289", "answer_1": "288", "answer_0": "315", "answer_3": "296"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p1": "0.150", "p0": "0.100", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.050", "p2": "0.150", "p0": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=918, unit_price=850, discount_pct=0, region=EU, sku_total=1768 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "779382", "v2": "78030", "v0": "780300", "v1": "1768"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = ops; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p2": "0.375", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p0": "0.100", "p3": "0.050", "p2": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.333", "p2": "0.667", "p3": "0.273"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed February 14, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 20", "date_2": "February 18", "date_1": "February 19", "date_3": "February 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.143", "p0": "0.250", "p1": "0.750", "p2": "0.750"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed August 22, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 01", "date_1": "August 30", "date_2": "September 04", "date_0": "September 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1874 per year. It increases by 15% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2123.1", "answer_3": "2110.1", "answer_2": "1592.9", "answer_1": "2155.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 488:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 5 grams of salt per serving and serves 5 people. A shopper bought a 25 gram box and a 250 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = $800; priority = high.\n\nClaim to verify: \"the date is $800\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Thursday, August 27, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "65", "answer_2": "30", "answer_3": "35", "answer_1": "37"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1202 per year. It increases by 25% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1502.5", "answer_0": "1482.5", "answer_3": "1498.5", "answer_2": "901.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.615", "p2": "0.615", "p0": "0.385", "p3": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Thursday, June 25, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1380 per year. It increases by 20% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1668", "answer_2": "1104", "answer_1": "1656", "answer_0": "1631"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1587 per year. It increases by 25% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1956.75", "answer_0": "1948.75", "answer_2": "1190.25", "answer_1": "1983.75"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = 2026-02-11; priority = low.\n\nClaim to verify: \"the date is 2026-02-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed December 27, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 30", "date_0": "December 30", "date_2": "December 28", "date_3": "January 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.667", "p2": "0.500", "p1": "0.833"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p1": "0.500", "p3": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1968. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "59", "age_0": "57", "age_1": "56", "age_2": "58"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1964. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "61", "age_3": "60", "age_0": "62", "age_2": "63"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=255, unit_price=592, discount_pct=0, region=EU, sku_total=847 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "150960", "v2": "15096", "v1": "847", "v3": "150705"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed April 26, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 01", "date_3": "May 01", "date_0": "April 29", "date_1": "April 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed September 01, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 10", "date_0": "September 10", "date_1": "September 08", "date_2": "September 08"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $210. A fee of $148 is charged, then a deposit of $56 arrives. A second fee of $74 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "53", "answer_1": "15", "answer_0": "44", "answer_3": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = 2026-10-10; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=990, unit_price=930, discount_pct=0, region=EU, sku_total=1920 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "92070", "v0": "920700", "v1": "1920", "v3": "921690"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 06:00 to 11:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "5h15", "dur_0": "4h45", "dur_3": "4h15", "dur_1": "5h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 432:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 5 grams of salt per serving and serves 12 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p2": "0.375", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $404. A fee of $100 is charged, then a deposit of $70 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "301", "answer_3": "294", "answer_1": "315", "answer_0": "324"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, February 09, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $336 per year. It increases by 10% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "302.4", "answer_0": "357.6", "answer_3": "389.6", "answer_1": "369.6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=800, unit_price=177, discount_pct=0, region=EU, sku_total=977 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "141600", "v3": "142400", "v2": "14160", "v1": "977"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed August 24, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 02", "date_3": "September 04", "date_2": "September 06", "date_0": "September 04"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p3": "0.167", "p0": "0.500", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 2 grams of salt per serving and serves 4 people. A shopper bought a 8 gram box and a 80 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1986. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "41", "age_1": "38", "age_2": "40", "age_0": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, March 22, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.625", "p3": "0.125", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = $775; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1965. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "62", "age_1": "60", "age_0": "61", "age_3": "63"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1277 per year. It increases by 10% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1149.3", "answer_0": "1370.7", "answer_3": "1360.7", "answer_1": "1404.7"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p3": "0.050", "p0": "0.150", "p2": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 231 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "37", "answer_0": "32", "answer_1": "13", "answer_2": "16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1984. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "40", "age_0": "41", "age_2": "42", "age_3": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=683, unit_price=242, discount_pct=0, region=EU, sku_total=925 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "16528", "v1": "925", "v3": "164603", "v0": "165286"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "77", "answer_1": "41", "answer_2": "35", "answer_3": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = 2026-11-04; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.467", "p2": "0.500", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $431. A fee of $84 is charged, then a deposit of $79 arrives. A second fee of $42 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "359", "answer_1": "364", "answer_2": "403", "answer_0": "384"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 360:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $16k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "16", "answer_2": "20", "answer_0": "36", "answer_1": "24"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = $400; priority = high.\n\nClaim to verify: \"the amount is $400\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_2": "28", "answer_1": "30", "answer_0": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed February 08, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 14", "date_3": "February 16", "date_0": "February 16", "date_1": "February 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "49", "answer_1": "35", "answer_2": "28", "answer_0": "77"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $403. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=653, unit_price=492, discount_pct=0, region=EU, sku_total=1145 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "32127", "v0": "321276", "v1": "1145", "v3": "321929"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.714", "p1": "0.714", "p0": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 09:00 to 21:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "12h15", "dur_0": "12h00", "dur_1": "12h45", "dur_3": "11h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed November 19, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 27", "date_2": "November 29", "date_0": "December 01", "date_3": "November 29"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 87 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "222", "answer_2": "239", "answer_0": "225", "answer_3": "220"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, April 19, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: status = open; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = 2026-10-11; priority = high.\n\nClaim to verify: \"the amount is 2026-10-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=932, unit_price=925, discount_pct=0, region=EU, sku_total=1857 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "86210", "v1": "1857", "v0": "862100", "v3": "863032"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $966 per year. It increases by 10% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1021.6", "answer_2": "869.4", "answer_0": "1046.6", "answer_1": "1062.6"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "77", "answer_3": "42", "answer_2": "35", "answer_1": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, October 02, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $439. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "27", "answer_0": "35", "answer_3": "10", "answer_2": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 11 people. A shopper bought a 22 gram box and a 220 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, September 05, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1967. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_1": "57", "age_3": "60", "age_2": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $734. A fee of $187 is charged, then a deposit of $34 arrives. A second fee of $93 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "493", "answer_0": "488", "answer_3": "495", "answer_1": "473"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=957, unit_price=497, discount_pct=0, region=EU, sku_total=1454 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "47562", "v1": "1454", "v0": "475629", "v3": "474672"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $558. A fee of $160 is charged, then a deposit of $82 arrives. A second fee of $80 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "412", "answer_3": "423", "answer_0": "400", "answer_2": "421"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1987. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "36", "age_2": "39", "age_1": "37", "age_0": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed February 27, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 09", "date_3": "March 11", "date_0": "March 13", "date_2": "March 15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = $179; priority = high.\n\nClaim to verify: \"the date is $179\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 318 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "92", "answer_1": "90", "answer_0": "108", "answer_2": "95"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed October 06, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 14", "date_1": "October 14", "date_0": "October 16", "date_3": "October 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 79 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "16", "answer_2": "14", "answer_1": "18", "answer_3": "31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "9", "answer_0": "15", "answer_2": "6", "answer_1": "9"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Contract signed June 06, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 14", "date_3": "June 16", "date_0": "June 17", "date_2": "June 15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, January 18, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1968. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "59", "age_3": "56", "age_1": "57", "age_0": "58"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.571", "p1": "0.571", "p3": "0.333", "p0": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1896 per year. It increases by 25% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2370", "answer_3": "2311", "answer_0": "2320", "answer_2": "1422"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.714", "p2": "0.714", "p3": "0.167", "p0": "0.286"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = $93; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 238:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $462 per year. It increases by 20% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "369.6", "answer_3": "529.4", "answer_1": "554.4", "answer_0": "512.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1968. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "55", "age_3": "58", "age_0": "56", "age_2": "57"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1616 per year. It increases by 25% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1212", "answer_0": "1970", "answer_3": "1938", "answer_1": "2020"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = 2026-10-14; priority = low.\n\nClaim to verify: \"the status is 2026-10-14\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1559 per year. It increases by 20% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1842.8", "answer_2": "1247.2", "answer_1": "1870.8", "answer_0": "1855.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=244, unit_price=578, discount_pct=0, region=EU, sku_total=822 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "14103", "v3": "140788", "v1": "822", "v0": "141032"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 888:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1988 per year. It increases by 15% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2285.2", "answer_1": "2286.2", "answer_2": "1689.8", "answer_0": "2250.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, October 12, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = $609; priority = high.\n\nClaim to verify: \"the owner is $609\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1968. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "56", "age_2": "57", "age_1": "55", "age_3": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $751. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.417", "p3": "0.545", "p0": "0.583", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = $277; priority = high.\n\nClaim to verify: \"the date is $277\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1466 per year. It increases by 15% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1685.9", "answer_0": "1656.9", "answer_2": "1246.1", "answer_3": "1679.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 8 grams of salt per serving and serves 10 people. A shopper bought a 80 gram box and a 800 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 982:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $426. A fee of $132 is charged, then a deposit of $41 arrives. A second fee of $66 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "269", "answer_2": "246", "answer_1": "259", "answer_3": "274"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.833", "p0": "0.667", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $221 per year. It increases by 5% this year, then a $40 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "192.05", "answer_2": "209.95", "answer_1": "232.05", "answer_3": "225.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = $733; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 746:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.571", "p3": "0.538", "p2": "0.429", "p1": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = $154; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $389 per year. It increases by 15% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "405.35", "answer_1": "447.35", "answer_2": "330.65", "answer_0": "413.35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, May 17, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = 2026-10-08; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $214. A fee of $156 is charged, then a deposit of $54 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "48", "answer_3": "24", "answer_1": "21", "answer_0": "34"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed February 25, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 05", "date_1": "March 03", "date_3": "March 05", "date_2": "March 07"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=472, unit_price=539, discount_pct=0, region=EU, sku_total=1011 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "254408", "v1": "1011", "v3": "254880", "v2": "25440"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1518 per year. It increases by 10% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1366.2", "answer_0": "1627.8", "answer_1": "1669.8", "answer_3": "1641.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=454, unit_price=261, discount_pct=0, region=EU, sku_total=715 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "715", "v2": "11849", "v0": "118494", "v3": "118040"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 341 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "72", "answer_0": "54", "answer_2": "48", "answer_3": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.778", "p0": "0.222", "p3": "0.125", "p1": "0.778"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = $156; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1867 per year. It increases by 15% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1586.95", "answer_3": "2144.05", "answer_1": "2147.05", "answer_0": "2108.05"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 10 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed April 23, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 29", "date_3": "April 29", "date_2": "May 01", "date_1": "April 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1544 per year. It increases by 10% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1389.6", "answer_3": "1703.4", "answer_1": "1698.4", "answer_0": "1675.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.125", "p2": "0.125", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $861. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed December 01, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 07", "date_1": "December 07", "date_3": "December 09", "date_0": "December 09"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed March 23, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 04", "date_0": "April 06", "date_3": "April 04", "date_1": "April 02"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed June 13, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 22", "date_1": "June 21", "date_3": "June 23", "date_0": "June 24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 3 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1630 per year. It increases by 15% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1385.5", "answer_1": "1874.5", "answer_3": "1844.5", "answer_0": "1853.5"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "44", "answer_3": "14", "answer_2": "42", "answer_0": "56"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $20k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $106. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 612:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.167", "p3": "0.083", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed October 12, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 21", "date_1": "October 19", "date_0": "October 21", "date_2": "October 19"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1623 per year. It increases by 20% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1902.6", "answer_2": "1298.4", "answer_1": "1947.6", "answer_0": "1927.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1964. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "62", "age_3": "60", "age_2": "63", "age_1": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1550 per year. It increases by 25% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1919.5", "answer_1": "1937.5", "answer_2": "1162.5", "answer_3": "1925.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $403 per year. It increases by 20% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "322.4", "answer_1": "483.6", "answer_3": "486.6", "answer_0": "466.6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.250", "p1": "0.375", "p2": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = 2026-12-07; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $586. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed April 04, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 11", "date_0": "April 10", "date_1": "April 09", "date_2": "April 12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 15:00 to 18:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h30", "dur_0": "3h15", "dur_3": "2h45", "dur_1": "3h45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = $557; priority = high.\n\nClaim to verify: \"the date is $557\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, February 03, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed April 22, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 30", "date_0": "May 04", "date_3": "May 02", "date_2": "May 06"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.667", "p2": "0.286", "p0": "0.714", "p1": "0.286"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $8k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 345:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed April 23, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 01", "date_0": "May 01", "date_2": "April 29", "date_1": "April 29"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1588 per year. It increases by 25% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1191", "answer_0": "1954", "answer_1": "1985", "answer_3": "1991"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 480:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 766:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed March 06, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 17", "date_2": "March 21", "date_1": "March 15", "date_0": "March 19"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed November 25, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 30", "date_0": "November 30", "date_2": "November 28", "date_1": "November 28"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1968 per year. It increases by 15% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2213.2", "answer_0": "2240.2", "answer_1": "2263.2", "answer_2": "1672.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p1": "0.500", "p3": "0.467"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.625", "p2": "0.375", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 99 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "276", "answer_2": "253", "answer_1": "242", "answer_0": "256"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $645. A fee of $168 is charged, then a deposit of $86 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "455", "answer_0": "479", "answer_3": "473", "answer_2": "449"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed May 25, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 28", "date_3": "May 30", "date_0": "May 28", "date_2": "May 26"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 278:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=288, unit_price=378, discount_pct=0, region=EU, sku_total=666 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "666", "v0": "108864", "v3": "109152", "v2": "10886"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 13:00 to 16:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "3h00", "dur_3": "2h15", "dur_2": "3h30", "dur_1": "3h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1985. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "41", "age_1": "38", "age_0": "39", "age_2": "40"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1987. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "40", "age_0": "39", "age_1": "38", "age_3": "41"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = 2026-11-13; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p1": "0.800", "p3": "0.111", "p2": "0.800"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.800", "p1": "0.800", "p3": "0.111", "p0": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1978. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "46", "age_1": "45", "age_3": "48", "age_2": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 379:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 388:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.467", "p2": "0.533", "p1": "0.533", "p3": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1964. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "59", "age_3": "58", "age_2": "61", "age_0": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 4 grams of salt per serving and serves 9 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed February 18, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 24", "date_0": "February 24", "date_1": "February 22", "date_2": "February 26"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Wednesday, March 25, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = closed; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = $336; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_1": "19", "answer_3": "28", "answer_2": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "9", "answer_3": "12", "answer_1": "13", "answer_0": "21"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $355. A fee of $194 is charged, then a deposit of $60 arrives. A second fee of $97 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "120", "answer_2": "131", "answer_0": "124", "answer_3": "116"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 125 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "118", "answer_0": "120", "answer_1": "137", "answer_3": "101"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 970:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = 2026-02-04; priority = high.\n\nClaim to verify: \"the status is 2026-02-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=613, unit_price=174, discount_pct=0, region=EU, sku_total=787 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "10666", "v3": "106049", "v1": "787", "v0": "106662"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 237:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 664:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, October 20, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1981. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "43", "age_1": "42", "age_3": "41", "age_2": "44"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Thursday, June 25, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = pending; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "46", "answer_2": "40", "answer_0": "88", "answer_3": "48"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 961:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 296 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "70", "answer_3": "72", "answer_2": "51", "answer_0": "54"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "30", "answer_1": "16", "answer_2": "10", "answer_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.167", "p0": "0.250", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, November 10, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed October 05, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "October 09", "date_2": "October 11", "date_3": "October 11", "date_1": "October 09"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = 2026-08-04; priority = high.\n\nClaim to verify: \"the date is 2026-08-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = 2026-07-27; priority = low.\n\nClaim to verify: \"the amount is 2026-07-27\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1964. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "60", "age_0": "61", "age_2": "62", "age_3": "63"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 5 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = 2026-02-02; priority = low.\n\nClaim to verify: \"the amount is 2026-02-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1988. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "37", "age_0": "38", "age_3": "40", "age_2": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 12 people. A shopper bought a 96 gram box and a 960 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $876 per year. It increases by 10% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "918.6", "answer_2": "788.4", "answer_1": "963.6", "answer_3": "893.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.467", "p3": "0.500", "p0": "0.533", "p1": "0.467"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.364", "p2": "0.636", "p3": "0.300", "p1": "0.636"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = $990; priority = high.\n\nClaim to verify: \"the date is $990\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $500. A fee of $164 is charged, then a deposit of $73 arrives. A second fee of $82 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "311", "answer_0": "327", "answer_1": "315", "answer_3": "346"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_0": "21", "answer_1": "15", "answer_3": "9"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Sunday, November 08, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1970. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "55", "age_3": "56", "age_0": "54", "age_1": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p1": "0.500", "p3": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 242 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_2": "39", "answer_1": "35", "answer_3": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1979. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "45", "age_2": "47", "age_3": "48", "age_0": "46"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.050", "p2": "0.150", "p0": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=874, unit_price=306, discount_pct=0, region=EU, sku_total=1180 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "267444", "v3": "268318", "v2": "26744", "v1": "1180"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "28", "answer_1": "23", "answer_2": "16", "answer_0": "44"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.333", "p3": "0.167", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "29", "answer_3": "20", "answer_2": "24", "answer_0": "44"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $771. A fee of $140 is charged, then a deposit of $68 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "636", "answer_0": "629", "answer_3": "599", "answer_1": "602"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $395. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "45", "answer_2": "15", "answer_1": "21", "answer_3": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 201 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "100", "answer_3": "108", "answer_1": "94", "answer_2": "95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = 2026-12-06; priority = high.\n\nClaim to verify: \"the owner is 2026-12-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1784 per year. It increases by 20% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1427.2", "answer_0": "2123.8", "answer_1": "2140.8", "answer_3": "2130.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $40k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p3": "0.429", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = 2026-04-01; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = $863; priority = high.\n\nClaim to verify: \"the amount is $863\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = 2026-10-16; priority = high.\n\nClaim to verify: \"the owner is 2026-10-16\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = 2026-12-02; priority = high.\n\nClaim to verify: \"the amount is 2026-12-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1387 per year. It increases by 5% this year, then a $49 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1407.35", "answer_1": "1456.35", "answer_2": "1317.65", "answer_3": "1423.35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $619 per year. It increases by 15% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "652.85", "answer_0": "677.85", "answer_2": "526.15", "answer_1": "711.85"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 10 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $355. A fee of $167 is charged, then a deposit of $32 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "137", "answer_2": "150", "answer_1": "110", "answer_3": "158"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed June 14, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 18", "date_0": "June 18", "date_2": "June 20", "date_3": "June 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=149, unit_price=878, discount_pct=0, region=EU, sku_total=1027 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1027", "v2": "13082", "v0": "130822", "v3": "130673"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "12", "answer_3": "9", "answer_2": "9", "answer_0": "18"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed April 03, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 09", "date_2": "April 11", "date_3": "April 09", "date_1": "April 07"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_3": "15", "answer_1": "28", "answer_2": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 10 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.833", "p3": "0.167", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 446:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.636", "p3": "0.300", "p0": "0.364", "p2": "0.636"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = $441; priority = high.\n\nClaim to verify: \"the status is $441\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed February 20, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 26", "date_0": "March 02", "date_3": "February 28", "date_2": "February 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1989. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "37", "age_3": "34", "age_0": "36", "age_1": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.500", "p1": "0.667", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1967. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "58", "age_2": "60", "age_0": "59", "age_3": "57"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1972. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "55", "age_3": "56", "age_1": "53", "age_0": "54"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = 2026-08-20; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $25k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 8 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 641:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 22:15 to 22:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "0h15", "dur_0": "-1h45", "dur_3": "-1h30", "dur_1": "0h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "30", "answer_2": "15", "answer_1": "18", "answer_3": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, April 28, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1150 per year. It increases by 25% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1423.5", "answer_1": "1437.5", "answer_2": "862.5", "answer_3": "1457.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $555. A fee of $143 is charged, then a deposit of $82 arrives. A second fee of $71 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "423", "answer_1": "436", "answer_2": "431", "answer_3": "406"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, July 19, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.167", "p0": "0.250", "p1": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p2": "0.050", "p1": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Friday, August 28, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 50 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "172", "answer_3": "169", "answer_2": "162", "answer_0": "160"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 334 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "87", "answer_2": "80", "answer_3": "81", "answer_0": "84"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.556", "p2": "0.444", "p3": "0.500", "p1": "0.444"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1758 per year. It increases by 15% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1988.7", "answer_3": "2026.7", "answer_2": "1494.3", "answer_1": "2021.7"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed November 26, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 10", "date_3": "December 08", "date_2": "December 08", "date_1": "December 06"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_0": "60", "answer_1": "30", "answer_3": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = 2026-06-07; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = 2026-02-26; priority = low.\n\nClaim to verify: \"the status is 2026-02-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.500", "p1": "0.625", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.500", "p3": "0.167", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.400", "p1": "0.500", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.667", "p2": "0.333", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = $89; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1983. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "43", "age_3": "44", "age_1": "41", "age_0": "42"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 607:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 267 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "288", "answer_1": "281", "answer_2": "306", "answer_3": "286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = 2026-09-18; priority = low.\n\nClaim to verify: \"the owner is 2026-09-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = 2026-05-11; priority = low.\n\nClaim to verify: \"the amount is 2026-05-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=196, unit_price=221, discount_pct=0, region=EU, sku_total=417 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "43512", "v0": "43316", "v2": "4331", "v1": "417"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.125", "p0": "0.375", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "56", "answer_0": "104", "answer_2": "48", "answer_1": "55"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p1": "0.625", "p0": "0.500", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $563 per year. It increases by 10% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "554.3", "answer_2": "506.7", "answer_0": "581.3", "answer_1": "619.3"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p2": "0.500", "p3": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p2": "0.375", "p0": "0.500", "p3": "0.125"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "36", "answer_3": "42", "answer_1": "43", "answer_0": "78"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = 2026-09-25; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 385:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, April 13, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 10:15 to 18:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "7h45", "dur_2": "8h15", "dur_3": "7h30", "dur_1": "8h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = $123; priority = low.\n\nClaim to verify: \"the amount is $123\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=185, unit_price=817, discount_pct=0, region=EU, sku_total=1002 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1002", "v2": "15114", "v0": "151145", "v3": "151330"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.500", "p2": "0.333", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 953:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 12 people. A shopper bought a 84 gram box and a 840 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 12:30 to 16:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "3h30", "dur_1": "3h45", "dur_2": "3h45", "dur_3": "2h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 7 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.250", "p0": "0.333", "p1": "0.417"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = $361; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = $486; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 523:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 6 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $635 per year. It increases by 10% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "698.5", "answer_0": "660.5", "answer_2": "571.5", "answer_3": "691.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_1": "18", "answer_3": "18", "answer_0": "30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $369. A fee of $58 is charged, then a deposit of $80 arrives. A second fee of $29 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "362", "answer_1": "342", "answer_3": "388", "answer_2": "351"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 7 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p3": "0.167", "p2": "0.500", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 248:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.556", "p0": "0.444", "p1": "0.556", "p3": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 266:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = ops; priority = low.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.375", "p3": "0.125", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "25", "answer_0": "40", "answer_3": "15", "answer_1": "28"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = $164; priority = high.\n\nClaim to verify: \"the date is $164\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Wednesday, May 27, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = $41; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 18:00 to 19:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "0h30", "dur_0": "1h15", "dur_1": "1h30", "dur_2": "1h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = $462; priority = low.\n\nClaim to verify: \"the amount is $462\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = $212; priority = low.\n\nClaim to verify: \"the owner is $212\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p3": "0.050", "p0": "0.100", "p2": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Wednesday, March 25, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Wednesday, May 27, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $325. A fee of $162 is charged, then a deposit of $80 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "174", "answer_2": "150", "answer_1": "182", "answer_0": "162"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.571", "p2": "0.429", "p1": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: status = 2026-01-08; priority = high.\n\nClaim to verify: \"the status is 2026-01-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 204:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.167", "p0": "0.250", "p1": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p2": "0.600", "p1": "0.600", "p0": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "19", "answer_2": "15", "answer_3": "12", "answer_0": "27"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, January 14, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1379 per year. It increases by 10% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1472.9", "answer_2": "1241.1", "answer_1": "1516.9", "answer_0": "1495.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $832 per year. It increases by 20% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "665.6", "answer_1": "998.4", "answer_0": "965.4", "answer_3": "949.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.417", "p2": "0.250", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = 2026-03-26; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, November 16, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 5 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.333", "p0": "0.667", "p3": "0.600"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.636", "p0": "0.667", "p1": "0.333", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 2 grams of salt per serving and serves 11 people. A shopper bought a 22 gram box and a 220 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $814. A fee of $108 is charged, then a deposit of $24 arrives. A second fee of $54 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "695", "answer_2": "656", "answer_0": "676", "answer_3": "706"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 766:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, April 11, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed April 11, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 17", "date_0": "April 16", "date_1": "April 15", "date_2": "April 18"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=380, unit_price=703, discount_pct=0, region=EU, sku_total=1083 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "267140", "v1": "1083", "v2": "26714", "v3": "266760"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, May 02, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, January 10, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p1": "0.375", "p3": "0.125", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $347. A fee of $39 is charged, then a deposit of $56 arrives. A second fee of $19 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "374", "answer_2": "355", "answer_0": "345", "answer_1": "323"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $66. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1971. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "51", "age_0": "53", "age_1": "52", "age_2": "54"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $418. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 283:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed June 20, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 27", "date_2": "June 28", "date_3": "June 29", "date_0": "June 30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $837 per year. It increases by 25% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "627.75", "answer_1": "1046.25", "answer_3": "1056.25", "answer_0": "1026.25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 499:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.400", "p0": "0.600", "p1": "0.400", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=186, unit_price=389, discount_pct=0, region=EU, sku_total=575 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "7235", "v0": "72354", "v3": "72540", "v1": "575"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.700", "p3": "0.667", "p2": "0.300", "p1": "0.300"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.625", "p0": "0.500", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1982. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "44", "age_3": "41", "age_0": "43", "age_1": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 4 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=671, unit_price=249, discount_pct=0, region=EU, sku_total=920 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "16707", "v0": "167079", "v3": "167750", "v1": "920"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = 2026-02-08; priority = low.\n\nClaim to verify: \"the date is 2026-02-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_0": "42", "answer_1": "30", "answer_2": "28"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $419 per year. It increases by 20% this year, then a $40 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "335.2", "answer_3": "497.8", "answer_0": "462.8", "answer_1": "502.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.100", "p1": "0.200", "p0": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "10", "answer_3": "25", "answer_0": "35", "answer_1": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=255, unit_price=587, discount_pct=0, region=EU, sku_total=842 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "14968", "v1": "842", "v3": "149430", "v0": "149685"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1982. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "43", "age_2": "44", "age_1": "42", "age_3": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 335 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "221", "answer_1": "224", "answer_2": "228", "answer_0": "240"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.667", "p3": "0.636", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 757:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 8 grams of salt per serving and serves 8 people. A shopper bought a 64 gram box and a 640 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed November 09, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 15", "date_3": "November 17", "date_0": "November 17", "date_2": "November 15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.429", "p1": "0.571", "p3": "0.385", "p2": "0.571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 131 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_3": "42", "answer_2": "25", "answer_1": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1979 per year. It increases by 25% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2459.75", "answer_1": "2473.75", "answer_2": "1484.25", "answer_3": "2472.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, June 23, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p0": "0.250", "p3": "0.125", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1990. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "36", "age_2": "37", "age_3": "34", "age_1": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "20", "answer_2": "18", "answer_3": "6", "answer_0": "24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1970. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "54", "age_1": "53", "age_3": "52", "age_2": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed August 23, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 01", "date_0": "September 03", "date_2": "September 05", "date_3": "September 03"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = $925; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1976. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "48", "age_0": "50", "age_2": "51", "age_1": "49"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed June 09, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 21", "date_0": "June 23", "date_2": "June 25", "date_1": "June 19"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $763. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_1": "23", "answer_2": "21", "answer_0": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, December 14, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Thursday, November 12, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 328 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "79", "answer_3": "103", "answer_1": "102", "answer_0": "96"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = 2026-06-08; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $669. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.200", "p1": "0.667", "p2": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, April 27, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Saturday, August 08, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = $187; priority = high.\n\nClaim to verify: \"the date is $187\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: owner = $951; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 606:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.111", "p1": "0.800", "p0": "0.200", "p2": "0.800"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 151 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "275", "answer_2": "263", "answer_3": "266", "answer_0": "270"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, January 20, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed April 17, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 29", "date_1": "April 27", "date_3": "April 29", "date_0": "May 01"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $334. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p2": "0.571", "p0": "0.429", "p3": "0.385"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.778", "p0": "0.222", "p1": "0.778"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_0": "64", "answer_2": "40", "answer_1": "43"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $757. A fee of $73 is charged, then a deposit of $82 arrives. A second fee of $36 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "701", "answer_2": "748", "answer_0": "730", "answer_3": "718"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, January 10, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = 2026-07-24; priority = low.\n\nClaim to verify: \"the date is 2026-07-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $840. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 8 grams of salt per serving and serves 8 people. A shopper bought a 64 gram box and a 640 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_0": "27", "answer_1": "21", "answer_3": "9"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.667", "p1": "0.833", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p3": "0.125", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 19:15 to 20:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "0h30", "dur_1": "0h45", "dur_3": "-1h45", "dur_0": "0h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.538", "p3": "0.417", "p0": "0.462", "p1": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=884, unit_price=635, discount_pct=0, region=EU, sku_total=1519 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1519", "v2": "56134", "v0": "561340", "v3": "562224"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=292, unit_price=957, discount_pct=0, region=EU, sku_total=1249 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "279736", "v0": "279444", "v2": "27944", "v1": "1249"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, June 17, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1961. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "64", "age_2": "66", "age_0": "65", "age_3": "67"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Thursday, December 10, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=251, unit_price=928, discount_pct=0, region=EU, sku_total=1179 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1179", "v3": "232677", "v2": "23292", "v0": "232928"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 5 grams of salt per serving and serves 12 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 7 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $821 per year. It increases by 5% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "845.05", "answer_2": "779.95", "answer_1": "862.05", "answer_3": "871.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "52", "answer_2": "48", "answer_0": "80", "answer_3": "32"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 3 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1965. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "63", "age_0": "61", "age_2": "62", "age_1": "60"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p3": "0.125", "p1": "0.375", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed February 05, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 13", "date_3": "February 13", "date_1": "February 11", "date_2": "February 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 793:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 219:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, April 16, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "15", "answer_2": "6", "answer_1": "11", "answer_0": "21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.636", "p3": "0.300", "p2": "0.636", "p0": "0.364"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $525. A fee of $187 is charged, then a deposit of $66 arrives. A second fee of $93 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "332", "answer_2": "333", "answer_3": "314", "answer_0": "311"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.538", "p2": "0.429", "p1": "0.429", "p0": "0.571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.333", "p0": "0.667", "p3": "0.636"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "51", "answer_3": "24", "answer_2": "48", "answer_0": "72"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "28", "answer_3": "16", "answer_1": "16", "answer_2": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $259. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.571", "p3": "0.500", "p1": "0.429", "p2": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1971. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "52", "age_2": "54", "age_3": "55", "age_0": "53"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1983. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "42", "age_2": "44", "age_3": "45", "age_0": "43"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = 2026-09-12; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 131 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "113", "answer_2": "122", "answer_0": "108", "answer_1": "123"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=917, unit_price=139, discount_pct=0, region=EU, sku_total=1056 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "12746", "v1": "1056", "v3": "128380", "v0": "127463"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $595. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=571, unit_price=320, discount_pct=0, region=EU, sku_total=891 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "18272", "v0": "182720", "v1": "891", "v3": "182149"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "39", "answer_3": "18", "answer_2": "36", "answer_0": "54"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $509. A fee of $91 is charged, then a deposit of $26 arrives. A second fee of $45 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "403", "answer_2": "409", "answer_1": "423", "answer_0": "399"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 863:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 8 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $502. A fee of $107 is charged, then a deposit of $38 arrives. A second fee of $53 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "359", "answer_0": "380", "answer_3": "383", "answer_2": "351"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $849. A fee of $166 is charged, then a deposit of $57 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "628", "answer_3": "681", "answer_1": "663", "answer_0": "657"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1974. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "51", "age_3": "50", "age_0": "52", "age_2": "53"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $590. A fee of $135 is charged, then a deposit of $36 arrives. A second fee of $67 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "414", "answer_0": "424", "answer_2": "421", "answer_3": "404"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 266 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "54", "answer_2": "57", "answer_1": "58", "answer_3": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.429", "p2": "0.533", "p1": "0.533", "p0": "0.467"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, April 11, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.100", "p1": "0.150", "p2": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.083", "p1": "0.417", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $217. A fee of $197 is charged, then a deposit of $29 arrives. A second fee of $98 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "-61", "answer_2": "-70", "answer_3": "-64", "answer_0": "-49"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=396, unit_price=609, discount_pct=0, region=EU, sku_total=1005 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "24116", "v3": "241560", "v1": "1005", "v0": "241164"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=157, unit_price=606, discount_pct=0, region=EU, sku_total=763 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "95299", "v2": "9514", "v0": "95142", "v1": "763"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 11 people. A shopper bought a 66 gram box and a 660 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed October 28, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 08", "date_0": "November 10", "date_1": "November 06", "date_2": "November 12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p1": "0.250", "p0": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $451 per year. It increases by 15% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "478.65", "answer_0": "507.65", "answer_1": "518.65", "answer_2": "383.35"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = 2026-03-23; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed June 11, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "June 19", "date_3": "June 19", "date_1": "June 17", "date_2": "June 17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 232 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_3": "44", "answer_1": "68", "answer_2": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $859. A fee of $59 is charged, then a deposit of $64 arrives. A second fee of $29 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "831", "answer_1": "843", "answer_3": "827", "answer_0": "835"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.500", "p1": "0.667", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $220. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1451 per year. It increases by 20% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1706.2", "answer_3": "1741.2", "answer_2": "1160.8", "answer_1": "1741.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1971. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "56", "age_1": "53", "age_2": "55", "age_0": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Saturday, December 12, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed October 17, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "October 22", "date_2": "October 20", "date_3": "October 23", "date_1": "October 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1804 per year. It increases by 20% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2164.8", "answer_2": "1443.2", "answer_0": "2146.8", "answer_3": "2134.8"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p3": "0.455", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.286", "p2": "0.625", "p0": "0.375", "p1": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $39k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = open; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 326 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "146", "answer_0": "144", "answer_3": "137", "answer_1": "149"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1751 per year. It increases by 25% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2145.75", "answer_2": "1313.25", "answer_0": "2176.75", "answer_1": "2188.75"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = 2026-05-22; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.500", "p0": "0.667", "p1": "0.833"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $855. A fee of $100 is charged, then a deposit of $37 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "742", "answer_1": "730", "answer_3": "723", "answer_2": "766"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = 2026-12-03; priority = low.\n\nClaim to verify: \"the status is 2026-12-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_2": "30", "answer_0": "72", "answer_1": "37"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $731. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Sunday, December 06, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = open; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed April 12, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 15", "date_2": "April 17", "date_1": "April 15", "date_3": "April 17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.400", "p2": "0.400", "p0": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 301 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "52", "answer_2": "43", "answer_1": "15", "answer_0": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = 2026-07-08; priority = low.\n\nClaim to verify: \"the owner is 2026-07-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.667", "p1": "0.833", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 201 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "141", "answer_2": "140", "answer_3": "151", "answer_0": "160"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p3": "0.462", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $230 per year. It increases by 5% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "212.5", "answer_3": "238.5", "answer_1": "241.5", "answer_2": "218.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 379:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed March 08, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 11", "date_3": "March 13", "date_1": "March 11", "date_2": "March 09"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed November 04, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 11", "date_2": "November 13", "date_1": "November 09", "date_0": "November 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $664. A fee of $84 is charged, then a deposit of $87 arrives. A second fee of $42 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "646", "answer_0": "625", "answer_2": "627", "answer_3": "637"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, December 24, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed March 16, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 21", "date_1": "March 19", "date_0": "March 19", "date_3": "March 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $772. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = $294; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p2": "0.083", "p1": "0.250", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed August 16, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 23", "date_2": "August 23", "date_0": "August 25", "date_3": "August 25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1962. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "64", "age_2": "65", "age_1": "63", "age_3": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = 2026-07-26; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p0": "0.200", "p1": "0.250", "p3": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Sunday, November 01, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $451. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $352. A fee of $121 is charged, then a deposit of $37 arrives. A second fee of $60 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "208", "answer_1": "179", "answer_3": "184", "answer_2": "190"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.467", "p3": "0.500", "p0": "0.533", "p1": "0.467"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed February 27, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 07", "date_1": "March 03", "date_3": "March 05", "date_0": "March 05"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Saturday, December 12, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, March 20, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $813. A fee of $44 is charged, then a deposit of $35 arrives. A second fee of $22 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "776", "answer_3": "795", "answer_1": "792", "answer_0": "782"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed May 11, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 22", "date_1": "May 18", "date_3": "May 20", "date_0": "May 20"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $616. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p0": "0.500", "p3": "0.125", "p1": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 06:45 to 18:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "10h15", "dur_2": "11h15", "dur_0": "11h00", "dur_1": "11h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=721, unit_price=953, discount_pct=0, region=EU, sku_total=1674 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "68711", "v1": "1674", "v0": "687113", "v3": "686392"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_3": "36", "answer_1": "36", "answer_0": "66"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_2": "32", "answer_0": "56", "answer_1": "35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed November 26, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 01", "date_3": "December 01", "date_1": "November 29", "date_2": "November 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $440 per year. It increases by 20% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "528", "answer_3": "445", "answer_2": "352", "answer_0": "480"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1973. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "52", "age_1": "50", "age_3": "49", "age_0": "51"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 297 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "45", "answer_0": "36", "answer_2": "47", "answer_3": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = open; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.125", "p1": "0.375", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $431. A fee of $166 is charged, then a deposit of $48 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "240", "answer_1": "218", "answer_3": "228", "answer_0": "230"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p3": "0.050", "p0": "0.200", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $784. A fee of $127 is charged, then a deposit of $65 arrives. A second fee of $63 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "682", "answer_0": "659", "answer_3": "683", "answer_2": "677"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1959. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "65", "age_2": "67", "age_3": "68", "age_0": "66"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = 2026-05-20; priority = high.\n\nClaim to verify: \"the owner is 2026-05-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, July 08, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.364", "p3": "0.600", "p0": "0.636", "p2": "0.364"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $554. A fee of $141 is charged, then a deposit of $85 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "400", "answer_2": "437", "answer_3": "453", "answer_0": "428"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_2": "9", "answer_1": "14", "answer_3": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 8 grams of salt per serving and serves 6 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = 2026-04-16; priority = high.\n\nClaim to verify: \"the amount is 2026-04-16\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=762, unit_price=712, discount_pct=0, region=EU, sku_total=1474 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "542544", "v1": "1474", "v2": "54254", "v3": "543306"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1963. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "62", "age_2": "63", "age_1": "61", "age_3": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed February 19, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 24", "date_2": "February 28", "date_0": "February 26", "date_3": "February 26"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 571:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.800", "p2": "0.800", "p0": "0.200", "p3": "0.111"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = $842; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed August 05, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 12", "date_2": "August 16", "date_0": "August 14", "date_3": "August 14"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=302, unit_price=869, discount_pct=0, region=EU, sku_total=1171 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "262438", "v3": "262136", "v2": "26243", "v1": "1171"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = 2026-11-07; priority = low.\n\nClaim to verify: \"the date is 2026-11-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed May 16, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 25", "date_0": "May 28", "date_3": "May 27", "date_2": "May 30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, February 10, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1947 per year. It increases by 25% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2391.75", "answer_1": "2433.75", "answer_0": "2403.75", "answer_2": "1460.25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.167", "p1": "0.500", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = $55; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 4 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = $344; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1130 per year. It increases by 25% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1387.5", "answer_0": "1394.5", "answer_2": "847.5", "answer_1": "1412.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = $303; priority = low.\n\nClaim to verify: \"the amount is $303\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1047 per year. It increases by 5% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1099.35", "answer_3": "1102.35", "answer_2": "994.65", "answer_0": "1086.35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $25k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1958. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "65", "age_0": "66", "age_3": "64", "age_2": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.833", "p0": "0.667", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 82 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "46", "answer_2": "45", "answer_1": "59", "answer_0": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=569, unit_price=940, discount_pct=0, region=EU, sku_total=1509 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "534860", "v3": "535429", "v1": "1509", "v2": "53486"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.083", "p1": "0.417", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1963. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "63", "age_2": "64", "age_1": "62", "age_3": "65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 283:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 940:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.545", "p2": "0.417", "p0": "0.583", "p1": "0.417"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 132 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "142", "answer_0": "144", "answer_3": "131", "answer_1": "139"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = 2026-10-21; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 845:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 07:15 to 19:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "11h30", "dur_3": "10h45", "dur_2": "11h45", "dur_1": "11h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, May 12, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=698, unit_price=346, discount_pct=0, region=EU, sku_total=1044 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "242206", "v2": "24150", "v1": "1044", "v0": "241508"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $468. A fee of $138 is charged, then a deposit of $79 arrives. A second fee of $69 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "370", "answer_2": "350", "answer_0": "340", "answer_3": "313"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.250", "p1": "0.417", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $692; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = open; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 12:45 to 20:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "7h15", "dur_1": "7h30", "dur_2": "7h45", "dur_3": "6h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.125", "p1": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "15", "answer_2": "8", "answer_3": "28", "answer_0": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed April 17, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 24", "date_2": "April 22", "date_0": "April 24", "date_1": "April 22"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.250", "p2": "0.167", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1987. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "38", "age_1": "37", "age_3": "40", "age_2": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=282, unit_price=885, discount_pct=0, region=EU, sku_total=1167 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "249288", "v2": "24957", "v0": "249570", "v1": "1167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1839 per year. It increases by 25% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1379.25", "answer_1": "2298.75", "answer_3": "2256.75", "answer_0": "2261.75"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 823:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $9k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=521, unit_price=480, discount_pct=0, region=EU, sku_total=1001 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "25008", "v0": "250080", "v1": "1001", "v3": "250601"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $337 per year. It increases by 20% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "404.4", "answer_3": "386.4", "answer_0": "372.4", "answer_2": "269.6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 07:45 to 15:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "6h30", "dur_1": "7h30", "dur_0": "7h15", "dur_2": "7h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $262. A fee of $158 is charged, then a deposit of $74 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "117", "answer_1": "128", "answer_0": "99", "answer_2": "102"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Friday, November 13, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.462", "p3": "0.500", "p0": "0.538", "p1": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $784 per year. It increases by 15% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "666.4", "answer_1": "901.6", "answer_0": "878.6", "answer_3": "839.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=292, unit_price=590, discount_pct=0, region=EU, sku_total=882 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "172572", "v2": "17228", "v1": "882", "v0": "172280"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, August 23, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = 2026-02-16; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, October 26, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed July 10, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 17", "date_2": "July 19", "date_0": "July 17", "date_1": "July 15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.462", "p0": "0.538", "p3": "0.500", "p1": "0.462"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = $285; priority = low.\n\nClaim to verify: \"the owner is $285\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $827. A fee of $42 is charged, then a deposit of $88 arrives. A second fee of $21 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "855", "answer_2": "861", "answer_1": "827", "answer_0": "852"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = $867; priority = low.\n\nClaim to verify: \"the date is $867\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p0": "0.500", "p3": "0.125", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed February 11, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 17", "date_1": "February 15", "date_0": "February 17", "date_2": "February 19"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed July 10, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 22", "date_3": "July 18", "date_0": "July 20", "date_1": "July 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = $370; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed September 18, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 29", "date_3": "September 27", "date_2": "September 27", "date_1": "September 25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, March 24, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = 2026-04-03; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1976. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "51", "age_1": "48", "age_2": "50", "age_0": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.462", "p1": "0.538", "p2": "0.538", "p3": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p3": "0.333", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.125", "p0": "0.250", "p1": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = $935; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed December 28, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 07", "date_2": "January 09", "date_0": "January 11", "date_3": "January 09"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = $342; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: amount = 2026-07-06; priority = high.\n\nClaim to verify: \"the amount is 2026-07-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $839. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "53", "age_0": "52", "age_1": "51", "age_3": "50"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 8 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $978 per year. It increases by 10% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1023.8", "answer_1": "1075.8", "answer_2": "880.2", "answer_0": "1036.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $401. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.333", "p3": "0.167", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $249. A fee of $158 is charged, then a deposit of $83 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "92", "answer_3": "92", "answer_0": "95", "answer_1": "93"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1717 per year. It increases by 20% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1373.6", "answer_3": "2031.4", "answer_1": "2060.4", "answer_0": "2013.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $496 per year. It increases by 15% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "558.4", "answer_2": "421.6", "answer_1": "570.4", "answer_0": "527.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.833", "p0": "0.667", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = $675; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 11 people. A shopper bought a 22 gram box and a 220 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 9 grams of salt per serving and serves 10 people. A shopper bought a 90 gram box and a 900 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "40", "answer_2": "32", "answer_1": "37", "answer_0": "72"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $842. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = 2026-04-04; priority = low.\n\nClaim to verify: \"the amount is 2026-04-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p3": "0.050", "p0": "0.100", "p2": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = 2026-02-28; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1969. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "55", "age_0": "56", "age_3": "58", "age_2": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=768, unit_price=746, discount_pct=0, region=EU, sku_total=1514 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "572160", "v1": "1514", "v0": "572928", "v2": "57292"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=177, unit_price=135, discount_pct=0, region=EU, sku_total=312 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "2389", "v3": "24072", "v1": "312", "v0": "23895"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed September 10, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 24", "date_1": "September 20", "date_3": "September 22", "date_2": "September 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.375", "p1": "0.500", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Saturday, March 21, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed June 15, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 23", "date_3": "June 25", "date_0": "June 25", "date_2": "June 27"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1972. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "51", "age_3": "50", "age_2": "53", "age_0": "52"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "77", "answer_3": "35", "answer_2": "42", "answer_1": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $451. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.150", "p1": "0.200", "p2": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 8 grams of salt per serving and serves 7 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 3 grams of salt per serving and serves 5 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=790, unit_price=288, discount_pct=0, region=EU, sku_total=1078 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "226730", "v2": "22752", "v1": "1078", "v0": "227520"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1982. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "43", "age_3": "40", "age_0": "42", "age_1": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 374 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "108", "answer_3": "99", "answer_1": "100", "answer_2": "90"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $11k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 314:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, May 28, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = open; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.083", "p2": "0.250", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 09:00 to 15:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "6h45", "dur_1": "6h45", "dur_3": "5h45", "dur_0": "6h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 119 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "128", "answer_0": "120", "answer_3": "111", "answer_1": "118"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $389. A fee of $39 is charged, then a deposit of $25 arrives. A second fee of $19 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "382", "answer_3": "341", "answer_1": "328", "answer_0": "356"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, October 06, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = 2026-10-26; priority = high.\n\nClaim to verify: \"the amount is 2026-10-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 784:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $378. A fee of $78 is charged, then a deposit of $58 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "335", "answer_2": "301", "answer_0": "319", "answer_3": "303"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 686:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.167", "p1": "0.250", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $244. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = $133; priority = low.\n\nClaim to verify: \"the status is $133\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1958. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "66", "age_2": "69", "age_0": "68", "age_1": "67"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = 2026-04-11; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 9 grams of salt per serving and serves 8 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed November 23, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 28", "date_2": "December 02", "date_3": "November 30", "date_0": "November 30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = $788; priority = low.\n\nClaim to verify: \"the date is $788\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 537:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p0": "0.250", "p3": "0.125", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 277 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "92", "answer_2": "94", "answer_0": "84", "answer_3": "81"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = $239; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.375", "p1": "0.500", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1001 per year. It increases by 15% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "850.85", "answer_3": "1122.15", "answer_0": "1104.15", "answer_1": "1151.15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed August 13, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 22", "date_3": "August 24", "date_0": "August 26", "date_2": "August 28"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = $105; priority = high.\n\nClaim to verify: \"the date is $105\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Thursday, April 09, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p2": "0.600", "p0": "0.400", "p3": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $953 per year. It increases by 25% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1191.25", "answer_0": "1164.25", "answer_2": "714.75", "answer_3": "1195.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $30k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p2": "0.600", "p1": "0.600", "p3": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $840. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $671. A fee of $190 is charged, then a deposit of $43 arrives. A second fee of $95 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "429", "answer_3": "401", "answer_1": "451", "answer_2": "446"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $795. A fee of $112 is charged, then a deposit of $41 arrives. A second fee of $56 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "665", "answer_1": "656", "answer_0": "668", "answer_2": "661"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 15:15 to 17:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "1h30", "dur_1": "2h30", "dur_0": "2h15", "dur_2": "2h45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = 2026-05-03; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1443 per year. It increases by 20% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1154.4", "answer_3": "1719.6", "answer_1": "1731.6", "answer_0": "1698.6"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = $540; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = ops; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $626. A fee of $165 is charged, then a deposit of $22 arrives. A second fee of $82 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "393", "answer_0": "401", "answer_1": "424", "answer_2": "414"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = $49; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1966. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "58", "age_2": "60", "age_3": "57", "age_0": "59"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = ops; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 7 grams of salt per serving and serves 12 people. A shopper bought a 84 gram box and a 840 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $721. A fee of $138 is charged, then a deposit of $70 arrives. A second fee of $69 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "584", "answer_3": "586", "answer_2": "556", "answer_1": "569"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Sunday, September 13, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=561, unit_price=833, discount_pct=0, region=EU, sku_total=1394 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "46731", "v3": "466752", "v1": "1394", "v0": "467313"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed November 17, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 01", "date_3": "November 29", "date_1": "November 27", "date_2": "November 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.400", "p0": "0.455", "p1": "0.545", "p2": "0.545"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1965. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "59", "age_2": "61", "age_3": "62", "age_0": "60"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=590, unit_price=796, discount_pct=0, region=EU, sku_total=1386 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "469050", "v0": "469640", "v2": "46964", "v1": "1386"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 319:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.545", "p0": "0.455", "p2": "0.545", "p3": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 539:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "35", "answer_3": "10", "answer_1": "27", "answer_2": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed August 20, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 28", "date_2": "September 03", "date_0": "September 01", "date_3": "August 30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 09:00 to 12:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "2h15", "dur_2": "3h00", "dur_0": "2h45", "dur_1": "3h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1549 per year. It increases by 25% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1936.25", "answer_0": "1924.25", "answer_2": "1161.75", "answer_3": "1895.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1404 per year. It increases by 25% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1053", "answer_0": "1732", "answer_1": "1755", "answer_3": "1714"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = 2026-09-06; priority = high.\n\nClaim to verify: \"the amount is 2026-09-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = $678; priority = low.\n\nClaim to verify: \"the amount is $678\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 8 grams of salt per serving and serves 9 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "10", "answer_1": "14", "answer_0": "30", "answer_3": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "8", "answer_0": "32", "answer_1": "14", "answer_3": "24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, December 27, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "35", "answer_2": "14", "answer_3": "21", "answer_1": "17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 196 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_2": "52", "answer_0": "42", "answer_1": "38"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.667", "p1": "0.333", "p3": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = 2026-12-20; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=468, unit_price=386, discount_pct=0, region=EU, sku_total=854 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "180648", "v1": "854", "v3": "180180", "v2": "18064"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p3": "0.333", "p2": "0.571", "p0": "0.429"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p2": "0.500", "p0": "0.667", "p3": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.727", "p0": "0.273", "p3": "0.200", "p2": "0.727"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 232 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_1": "28", "answer_3": "36", "answer_0": "30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $636 per year. It increases by 25% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "477", "answer_1": "795", "answer_0": "747", "answer_3": "732"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 12 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $249 per year. It increases by 15% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "211.65", "answer_1": "286.35", "answer_3": "238.35", "answer_0": "266.35"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 12 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=534, unit_price=637, discount_pct=0, region=EU, sku_total=1171 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "34015", "v0": "340158", "v3": "339624", "v1": "1171"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = closed; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1969. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "53", "age_1": "54", "age_0": "55", "age_2": "56"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $808. A fee of $139 is charged, then a deposit of $39 arrives. A second fee of $69 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "665", "answer_0": "639", "answer_2": "616", "answer_3": "624"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 8 grams of salt per serving and serves 7 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.385", "p1": "0.385", "p3": "0.583", "p0": "0.615"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=853, unit_price=753, discount_pct=0, region=EU, sku_total=1606 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1606", "v0": "642309", "v2": "64230", "v3": "641456"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = ops; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_1": "38", "answer_2": "36", "answer_3": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = 2026-08-18; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, May 23, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $295 per year. It increases by 10% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "265.5", "answer_1": "324.5", "answer_0": "298.5", "answer_3": "264.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.150", "p2": "0.050", "p0": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 257 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "67", "answer_3": "73", "answer_2": "59", "answer_0": "63"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=110, unit_price=801, discount_pct=0, region=EU, sku_total=911 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "8811", "v3": "88220", "v0": "88110", "v1": "911"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1974. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "52", "age_1": "50", "age_0": "51", "age_3": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 11:15 to 11:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "0h30", "dur_3": "-1h30", "dur_0": "0h00", "dur_1": "0h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 169 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "87", "answer_3": "74", "answer_0": "80", "answer_1": "95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 8 grams of salt per serving and serves 3 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed August 21, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 24", "date_2": "August 24", "date_0": "August 26", "date_3": "August 26"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, January 12, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 830:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 108 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "117", "answer_3": "146", "answer_2": "130", "answer_0": "128"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $420. A fee of $65 is charged, then a deposit of $47 arrives. A second fee of $32 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "399", "answer_0": "370", "answer_1": "367", "answer_3": "388"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed April 22, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 27", "date_1": "April 25", "date_2": "April 29", "date_3": "April 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $860. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed November 02, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 07", "date_0": "November 05", "date_1": "November 05", "date_2": "November 03"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=444, unit_price=272, discount_pct=0, region=EU, sku_total=716 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "120324", "v2": "12076", "v1": "716", "v0": "120768"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 13:30 to 20:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "6h30", "dur_3": "5h30", "dur_0": "5h45", "dur_2": "6h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $453. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 210 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "78", "answer_3": "78", "answer_0": "80", "answer_1": "94"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed March 28, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 02", "date_2": "April 03", "date_0": "April 01", "date_1": "March 31"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed March 03, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 07", "date_0": "March 09", "date_1": "March 07", "date_3": "March 09"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed October 14, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 24", "date_0": "October 28", "date_2": "October 30", "date_3": "October 26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $9k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1414 per year. It increases by 25% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1730.5", "answer_3": "1758.5", "answer_2": "1060.5", "answer_1": "1767.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $978 per year. It increases by 10% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "880.2", "answer_1": "1075.8", "answer_3": "1050.8", "answer_0": "1044.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p1": "0.500", "p3": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 10:15 to 21:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "10h15", "dur_2": "11h15", "dur_1": "11h15", "dur_0": "11h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 652:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = 2026-06-24; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=267, unit_price=299, discount_pct=0, region=EU, sku_total=566 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "7983", "v0": "79833", "v1": "566", "v3": "79566"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Friday, February 06, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = $203; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 839:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $21k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $522. A fee of $57 is charged, then a deposit of $39 arrives. A second fee of $28 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "476", "answer_2": "487", "answer_1": "464", "answer_3": "486"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 344:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.050", "p2": "0.150", "p0": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, April 20, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 06:30 to 12:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "6h15", "dur_0": "6h00", "dur_3": "5h15", "dur_2": "6h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = open; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $13k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 13:15 to 19:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "6h15", "dur_1": "6h00", "dur_0": "5h45", "dur_3": "5h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 08:30 to 09:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "0h15", "dur_1": "0h45", "dur_2": "0h45", "dur_3": "-1h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p2": "0.250", "p3": "0.083", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 189 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_1": "34", "answer_0": "18", "answer_3": "4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1986. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "38", "age_3": "36", "age_1": "37", "age_2": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, February 21, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1958. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "68", "age_0": "67", "age_3": "65", "age_1": "66"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $848. A fee of $113 is charged, then a deposit of $23 arrives. A second fee of $56 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "702", "answer_2": "720", "answer_1": "704", "answer_3": "689"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_1": "29", "answer_0": "44", "answer_3": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = 2026-06-10; priority = low.\n\nClaim to verify: \"the status is 2026-06-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $306. A fee of $129 is charged, then a deposit of $63 arrives. A second fee of $64 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "147", "answer_1": "147", "answer_3": "186", "answer_0": "176"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, March 06, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Tuesday, November 24, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = 2026-09-01; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p3": "0.125", "p1": "0.500", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 3 grams of salt per serving and serves 3 people. A shopper bought a 9 gram box and a 90 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = open; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 211 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_1": "46", "answer_2": "44", "answer_3": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 4 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = 2026-02-10; priority = low.\n\nClaim to verify: \"the owner is 2026-02-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = ops; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $330. A fee of $67 is charged, then a deposit of $78 arrives. A second fee of $33 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "294", "answer_1": "278", "answer_0": "308", "answer_2": "318"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1972. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "51", "age_3": "50", "age_2": "53", "age_0": "52"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = $20; priority = high.\n\nClaim to verify: \"the status is $20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $4k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 14:15 to 18:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "3h00", "dur_2": "3h45", "dur_1": "4h00", "dur_0": "3h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 7 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = $704; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $594. A fee of $169 is charged, then a deposit of $61 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "416", "answer_2": "414", "answer_3": "398", "answer_0": "402"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = $857; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1962. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "64", "age_3": "61", "age_0": "63", "age_1": "62"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_1": "23", "answer_2": "18", "answer_3": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed September 13, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 18", "date_3": "September 20", "date_1": "September 18", "date_2": "September 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1958. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "66", "age_3": "64", "age_2": "67", "age_1": "65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $847. A fee of $87 is charged, then a deposit of $25 arrives. A second fee of $43 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "747", "answer_0": "742", "answer_2": "735", "answer_1": "752"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=964, unit_price=972, discount_pct=0, region=EU, sku_total=1936 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1936", "v2": "93700", "v3": "937972", "v0": "937008"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = $662; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=722, unit_price=623, discount_pct=0, region=EU, sku_total=1345 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "450528", "v2": "44980", "v0": "449806", "v1": "1345"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=226, unit_price=991, discount_pct=0, region=EU, sku_total=1217 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "223966", "v2": "22396", "v1": "1217", "v3": "223740"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $20k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "16", "answer_2": "20", "answer_0": "36", "answer_1": "24"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = ops; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = $962; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 20:15 to 21:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h00", "dur_1": "0h45", "dur_3": "-1h45", "dur_0": "0h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "16", "answer_3": "8", "answer_2": "8", "answer_1": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 504:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1238 per year. It increases by 10% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1298.8", "answer_1": "1361.8", "answer_2": "1114.2", "answer_0": "1334.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $439. A fee of $179 is charged, then a deposit of $61 arrives. A second fee of $89 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "203", "answer_2": "237", "answer_0": "232", "answer_1": "248"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 531:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 08:30 to 16:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "7h30", "dur_1": "7h45", "dur_3": "6h45", "dur_0": "7h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed June 04, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "June 09", "date_2": "June 11", "date_3": "June 09", "date_1": "June 07"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $665. A fee of $55 is charged, then a deposit of $37 arrives. A second fee of $27 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "625", "answer_0": "620", "answer_1": "646", "answer_2": "591"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_3": "12", "answer_1": "15", "answer_2": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 07:45 to 12:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "3h30", "dur_0": "3h45", "dur_2": "4h00", "dur_1": "4h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = ops; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.167", "p1": "0.250", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $393. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.250", "p3": "0.083", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 987:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 600:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1962. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "63", "age_0": "62", "age_3": "60", "age_1": "61"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 12:45 to 18:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "4h45", "dur_1": "5h45", "dur_0": "5h30", "dur_2": "5h45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $678. A fee of $67 is charged, then a deposit of $77 arrives. A second fee of $33 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "678", "answer_3": "637", "answer_2": "630", "answer_0": "655"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.714", "p0": "0.286", "p3": "0.167", "p2": "0.714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed February 02, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 16", "date_1": "February 12", "date_3": "February 14", "date_2": "February 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.125", "p3": "0.125", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1970. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "56", "age_0": "55", "age_3": "57", "age_1": "54"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "51", "age_2": "53", "age_0": "52", "age_3": "50"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=269, unit_price=803, discount_pct=0, region=EU, sku_total=1072 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "216007", "v2": "21600", "v1": "1072", "v3": "215738"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = $625; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.250", "p1": "0.417", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = $773; priority = high.\n\nClaim to verify: \"the date is $773\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed June 06, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 15", "date_0": "June 16", "date_2": "June 18", "date_1": "June 13"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p0": "0.250", "p1": "0.375", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed March 11, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 20", "date_3": "March 22", "date_0": "March 24", "date_2": "March 22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "22", "answer_2": "15", "answer_3": "21", "answer_0": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = 2026-03-05; priority = low.\n\nClaim to verify: \"the owner is 2026-03-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $476. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $638. A fee of $163 is charged, then a deposit of $20 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "423", "answer_1": "416", "answer_2": "406", "answer_0": "414"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1643 per year. It increases by 15% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1889.45", "answer_2": "1396.55", "answer_3": "1852.45", "answer_0": "1865.45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed April 28, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 04", "date_3": "May 06", "date_2": "May 04", "date_0": "May 06"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 9 grams of salt per serving and serves 6 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, January 17, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.429", "p3": "0.500", "p1": "0.429", "p0": "0.571"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, March 16, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = pending; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 978:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $13k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = $85; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1757 per year. It increases by 25% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2179.25", "answer_1": "2196.25", "answer_2": "1317.75", "answer_3": "2155.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $859. A fee of $32 is charged, then a deposit of $38 arrives. A second fee of $16 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "849", "answer_3": "866", "answer_1": "866", "answer_2": "858"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 129 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "47", "answer_0": "60", "answer_2": "62", "answer_1": "48"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 783:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 369:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "21", "answer_0": "35", "answer_3": "14", "answer_1": "23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 09:15 to 15:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "5h45", "dur_2": "5h30", "dur_3": "4h45", "dur_0": "5h00"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.500", "p2": "0.333", "p1": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, September 12, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = open; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.222", "p0": "0.300", "p2": "0.700", "p1": "0.700"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.300", "p0": "0.700", "p2": "0.300", "p3": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p0": "0.400", "p1": "0.600", "p2": "0.600"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p1": "0.200", "p3": "0.050", "p2": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 185 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "9", "answer_0": "12", "answer_2": "23", "answer_3": "-4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.286", "p1": "0.714", "p3": "0.167", "p2": "0.714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed January 19, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 24", "date_0": "January 26", "date_1": "January 24", "date_3": "January 26"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.150", "p2": "0.050", "p0": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1972. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "52", "age_2": "54", "age_3": "55", "age_0": "53"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 225 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "89", "answer_1": "91", "answer_2": "52", "answer_0": "72"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 551:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = 2026-02-17; priority = low.\n\nClaim to verify: \"the owner is 2026-02-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = $798; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1978. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "48", "age_2": "49", "age_3": "46", "age_1": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed March 12, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 19", "date_1": "March 17", "date_2": "March 21", "date_0": "March 19"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $545. A fee of $112 is charged, then a deposit of $78 arrives. A second fee of $56 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "472", "answer_1": "426", "answer_0": "455", "answer_3": "459"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 458:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 581:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, August 17, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $321 per year. It increases by 5% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "304.95", "answer_1": "337.05", "answer_3": "274.05", "answer_0": "289.05"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $12k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "9", "answer_0": "21", "answer_3": "12", "answer_1": "13"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = $771; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.583", "p3": "0.545", "p1": "0.417", "p2": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $18k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = ops; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = $432; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = ops; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 9 grams of salt per serving and serves 7 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1982. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "44", "age_2": "43", "age_0": "42", "age_1": "41"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed November 26, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 04", "date_2": "December 06", "date_3": "December 04", "date_1": "December 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = $509; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 143 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "89", "answer_0": "100", "answer_1": "81", "answer_2": "82"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $882 per year. It increases by 20% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1047.4", "answer_1": "1058.4", "answer_0": "1039.4", "answer_2": "705.6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = $214; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.150", "p2": "0.050", "p0": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 8 grams of salt per serving and serves 9 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = $757; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 692:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 788:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = 2026-01-08; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 7 people. A shopper bought a 14 gram box and a 140 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "39", "answer_0": "88", "answer_2": "32", "answer_3": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = 2026-03-24; priority = high.\n\nClaim to verify: \"the date is 2026-03-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $482. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 208:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.636", "p1": "0.333", "p2": "0.333", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Sunday, June 21, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed October 18, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "October 28", "date_1": "October 26", "date_2": "October 30", "date_3": "October 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $413. A fee of $172 is charged, then a deposit of $59 arrives. A second fee of $86 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "214", "answer_3": "240", "answer_2": "240", "answer_1": "217"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $829. A fee of $100 is charged, then a deposit of $61 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "748", "answer_3": "713", "answer_0": "740", "answer_1": "725"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 4 grams of salt per serving and serves 9 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = $422; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_0": "24", "answer_2": "6", "answer_1": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 352 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "126", "answer_1": "107", "answer_3": "114", "answer_0": "120"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed June 27, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 01", "date_2": "June 30", "date_3": "July 03", "date_0": "July 02"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed February 01, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 12", "date_3": "February 10", "date_1": "February 08", "date_0": "February 10"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed August 02, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 06", "date_0": "August 06", "date_3": "August 08", "date_2": "August 08"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=435, unit_price=609, discount_pct=0, region=EU, sku_total=1044 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1044", "v0": "264915", "v2": "26491", "v3": "264480"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p3": "0.050", "p2": "0.050", "p0": "0.100"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, October 09, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $357 per year. It increases by 10% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "327.7", "answer_2": "321.3", "answer_0": "361.7", "answer_1": "392.7"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p1": "0.375", "p0": "0.625", "p3": "0.571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "8", "answer_1": "14", "answer_2": "12", "answer_0": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $450. A fee of $131 is charged, then a deposit of $50 arrives. A second fee of $65 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "280", "answer_1": "334", "answer_0": "304", "answer_2": "278"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=462, unit_price=917, discount_pct=0, region=EU, sku_total=1379 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "423654", "v3": "424116", "v1": "1379", "v2": "42365"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p2": "0.600", "p1": "0.600", "p3": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.545", "p0": "0.455", "p3": "0.400", "p1": "0.545"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, January 12, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = 2026-03-22; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.250", "p3": "0.083", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.714", "p2": "0.250", "p0": "0.750"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = open; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: date = 2026-03-27; priority = low.\n\nClaim to verify: \"the date is 2026-03-27\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = $273; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $895. A fee of $105 is charged, then a deposit of $74 arrives. A second fee of $52 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "794", "answer_0": "812", "answer_3": "794", "answer_2": "803"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = $871; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Sunday, March 15, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed April 10, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 18", "date_0": "April 22", "date_2": "April 20", "date_3": "April 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_3": "24", "answer_1": "40", "answer_2": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p3": "0.083", "p2": "0.167", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 7 grams of salt per serving and serves 5 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = 2026-07-09; priority = low.\n\nClaim to verify: \"the date is 2026-07-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "20", "answer_0": "32", "answer_1": "23", "answer_3": "12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $176. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.333", "p0": "0.250", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_0": "24", "answer_1": "18", "answer_3": "8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = 2026-03-16; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = $910; priority = high.\n\nClaim to verify: \"the amount is $910\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1981. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "45", "age_3": "42", "age_1": "43", "age_0": "44"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = 2026-09-13; priority = high.\n\nClaim to verify: \"the owner is 2026-09-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 13:15 to 19:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "6h15", "dur_3": "5h30", "dur_2": "6h30", "dur_1": "6h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Wednesday, August 12, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=364, unit_price=221, discount_pct=0, region=EU, sku_total=585 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "585", "v0": "80444", "v2": "8044", "v3": "80080"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 3 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.250", "p2": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=508, unit_price=340, discount_pct=0, region=EU, sku_total=848 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "173228", "v0": "172720", "v1": "848", "v2": "17272"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "30", "answer_0": "60", "answer_1": "35", "answer_2": "30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, March 17, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p2": "0.050", "p1": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1983. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "45", "age_2": "44", "age_1": "42", "age_0": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = $69; priority = low.\n\nClaim to verify: \"the date is $69\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p3": "0.125", "p2": "0.125", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Wednesday, October 07, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.444", "p0": "0.500", "p2": "0.500", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = $722; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Wednesday, November 18, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 251:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 2 grams of salt per serving and serves 11 people. A shopper bought a 22 gram box and a 220 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p1": "0.500", "p3": "0.467"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed June 15, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 20", "date_3": "June 22", "date_0": "June 22", "date_2": "June 24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 835:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.462", "p0": "0.538", "p3": "0.500", "p2": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $842. A fee of $80 is charged, then a deposit of $58 arrives. A second fee of $40 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "796", "answer_1": "777", "answer_0": "780", "answer_3": "785"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $547; priority = high.\n\nClaim to verify: \"the owner is $547\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = closed; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed February 21, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 24", "date_3": "February 27", "date_1": "February 25", "date_0": "February 26"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1972. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "52", "age_1": "51", "age_2": "53", "age_3": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed December 23, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 01", "date_3": "December 30", "date_0": "December 30", "date_1": "December 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=188, unit_price=650, discount_pct=0, region=EU, sku_total=838 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "122012", "v1": "838", "v2": "12220", "v0": "122200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1677 per year. It increases by 25% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1257.75", "answer_3": "2087.25", "answer_0": "2073.25", "answer_1": "2096.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 431:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.533", "p3": "0.429", "p2": "0.533", "p0": "0.467"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, November 28, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $582 per year. It increases by 25% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "719.5", "answer_1": "727.5", "answer_0": "700.5", "answer_2": "436.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $878. A fee of $156 is charged, then a deposit of $53 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "722", "answer_0": "697", "answer_2": "672", "answer_3": "701"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = 2026-04-22; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = $321; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.250", "p2": "0.150", "p0": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $675. A fee of $62 is charged, then a deposit of $74 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "650", "answer_3": "675", "answer_0": "656", "answer_2": "635"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 629:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 8 grams of salt per serving and serves 9 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Tuesday, April 28, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $748 per year. It increases by 15% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "795.2", "answer_0": "834.2", "answer_2": "635.8", "answer_1": "860.2"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1972. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "54", "age_1": "51", "age_2": "53", "age_0": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 56 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "195", "answer_0": "192", "answer_1": "175", "answer_3": "174"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 10:45 to 18:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "7h30", "dur_0": "7h00", "dur_2": "7h15", "dur_3": "6h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_2": "30", "answer_1": "33", "answer_0": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $292. A fee of $156 is charged, then a deposit of $25 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "74", "answer_0": "83", "answer_2": "110", "answer_3": "88"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = $880; priority = high.\n\nClaim to verify: \"the amount is $880\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed February 24, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 28", "date_3": "March 02", "date_0": "March 02", "date_2": "February 28"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = $13; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 198 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "26", "answer_0": "32", "answer_1": "43", "answer_3": "51"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $412. A fee of $101 is charged, then a deposit of $37 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "298", "answer_1": "278", "answer_2": "320", "answer_3": "309"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p2": "0.625", "p0": "0.375", "p3": "0.286"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.625", "p3": "0.125", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 139 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "58", "answer_3": "56", "answer_0": "63", "answer_1": "69"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $328. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.125", "p1": "0.625", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.429", "p1": "0.571", "p2": "0.571", "p3": "0.385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $325. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p2": "0.500", "p3": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 276 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "140", "answer_2": "135", "answer_3": "130", "answer_0": "120"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1959. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "67", "age_2": "68", "age_3": "69", "age_1": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $309. A fee of $128 is charged, then a deposit of $79 arrives. A second fee of $64 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "196", "answer_2": "206", "answer_1": "208", "answer_3": "218"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 454:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $591 per year. It increases by 20% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "694.2", "answer_2": "472.8", "answer_1": "709.2", "answer_3": "726.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1968. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "58", "age_0": "57", "age_1": "56", "age_3": "59"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $629. A fee of $75 is charged, then a deposit of $67 arrives. A second fee of $37 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "600", "answer_1": "588", "answer_0": "584", "answer_2": "588"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 12:30 to 22:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "9h45", "dur_1": "10h00", "dur_3": "9h00", "dur_2": "10h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=769, unit_price=334, discount_pct=0, region=EU, sku_total=1103 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "25684", "v3": "257615", "v1": "1103", "v0": "256846"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed September 11, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 17", "date_1": "September 15", "date_0": "September 17", "date_2": "September 19"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 5 grams of salt per serving and serves 10 people. A shopper bought a 50 gram box and a 500 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = $849; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $564 per year. It increases by 25% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "705", "answer_2": "423", "answer_0": "676", "answer_1": "705"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "45", "answer_3": "30", "answer_1": "21", "answer_2": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 7 people. A shopper bought a 14 gram box and a 140 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 628:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $10k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $791. A fee of $61 is charged, then a deposit of $37 arrives. A second fee of $30 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "755", "answer_3": "725", "answer_1": "722", "answer_0": "737"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = 2026-06-18; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $3k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = 2026-08-04; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 257:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p2": "0.571", "p0": "0.429", "p3": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "9", "answer_3": "9", "answer_0": "15", "answer_2": "6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1090 per year. It increases by 25% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1362.5", "answer_2": "817.5", "answer_3": "1366.5", "answer_0": "1336.5"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, July 08, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 225 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "288", "answer_3": "299", "answer_2": "286", "answer_1": "308"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1799 per year. It increases by 5% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1709.05", "answer_0": "1863.95", "answer_1": "1888.95", "answer_3": "1878.95"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 207:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = 2026-10-07; priority = high.\n\nClaim to verify: \"the date is 2026-10-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1974. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "52", "age_2": "53", "age_3": "54", "age_1": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = 2026-01-28; priority = high.\n\nClaim to verify: \"the amount is 2026-01-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_0": "55", "answer_1": "35", "answer_3": "25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed June 26, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 07", "date_0": "July 09", "date_2": "July 11", "date_1": "July 05"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p2": "0.375", "p3": "0.571", "p0": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $396 per year. It increases by 5% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "386.8", "answer_3": "394.8", "answer_1": "415.8", "answer_2": "376.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.222", "p1": "0.700", "p0": "0.300", "p2": "0.700"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 8 grams of salt per serving and serves 6 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p3": "0.050", "p2": "0.100", "p0": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $773 per year. It increases by 15% this year, then a $40 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "888.95", "answer_2": "657.05", "answer_3": "886.95", "answer_0": "848.95"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = closed; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1621 per year. It increases by 15% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1864.15", "answer_0": "1848.15", "answer_3": "1882.15", "answer_2": "1377.85"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $2k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, August 15, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=371, unit_price=133, discount_pct=0, region=EU, sku_total=504 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "49343", "v3": "48972", "v2": "4934", "v1": "504"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 15:00 to 18:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h15", "dur_3": "2h30", "dur_1": "3h30", "dur_0": "2h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "17", "answer_3": "30", "answer_0": "42", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, February 12, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1961. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "63", "age_3": "65", "age_1": "62", "age_2": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=275, unit_price=304, discount_pct=0, region=EU, sku_total=579 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "83600", "v2": "8360", "v3": "83875", "v1": "579"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 08:00 to 17:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "9h00", "dur_1": "9h15", "dur_3": "8h15", "dur_0": "8h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: status = $115; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed August 08, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 20", "date_1": "August 18", "date_2": "August 23", "date_0": "August 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = 2026-07-01; priority = low.\n\nClaim to verify: \"the owner is 2026-07-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $893. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "39", "answer_3": "18", "answer_0": "54", "answer_2": "36"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 865:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed January 26, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 29", "date_3": "January 31", "date_1": "January 29", "date_2": "January 31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "6", "answer_1": "10", "answer_3": "12", "answer_0": "18"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $278. A fee of $196 is charged, then a deposit of $50 arrives. A second fee of $98 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "34", "answer_1": "41", "answer_3": "51", "answer_2": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, October 07, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed December 07, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 17", "date_3": "December 15", "date_1": "December 13", "date_0": "December 15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = 2026-01-25; priority = low.\n\nClaim to verify: \"the date is 2026-01-25\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $480 per year. It increases by 25% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "360", "answer_0": "566", "answer_3": "527", "answer_1": "600"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed May 02, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 12", "date_3": "May 11", "date_1": "May 09", "date_2": "May 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p1": "0.500", "p2": "0.250", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $386. A fee of $68 is charged, then a deposit of $70 arrives. A second fee of $34 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "377", "answer_0": "354", "answer_3": "350", "answer_2": "337"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1712 per year. It increases by 5% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1775.6", "answer_2": "1626.4", "answer_1": "1797.6", "answer_3": "1810.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1969. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "57", "age_1": "54", "age_0": "55", "age_2": "56"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=566, unit_price=946, discount_pct=0, region=EU, sku_total=1512 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1512", "v0": "535436", "v2": "53543", "v3": "536002"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p3": "0.050", "p1": "0.250", "p0": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 776:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, April 21, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.364", "p1": "0.583", "p2": "0.583", "p0": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $277. A fee of $192 is charged, then a deposit of $40 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "48", "answer_3": "1", "answer_1": "10", "answer_0": "29"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 12 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_0": "40", "answer_1": "22", "answer_2": "16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1969. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "56", "age_1": "54", "age_0": "55", "age_3": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = $907; priority = low.\n\nClaim to verify: \"the date is $907\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_1": "16", "answer_0": "36", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=613, unit_price=110, discount_pct=0, region=EU, sku_total=723 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "68043", "v2": "6743", "v1": "723", "v0": "67430"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.444", "p0": "0.556", "p1": "0.444"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = $90; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 5 grams of salt per serving and serves 10 people. A shopper bought a 50 gram box and a 500 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=596, unit_price=177, discount_pct=0, region=EU, sku_total=773 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "105492", "v1": "773", "v3": "106088", "v2": "10549"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.333", "p1": "0.500", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 239 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "278", "answer_2": "299", "answer_0": "280", "answer_1": "267"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Friday, May 22, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $802 per year. It increases by 20% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "939.4", "answer_1": "962.4", "answer_3": "979.4", "answer_2": "641.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = 2026-07-06; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1968. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "57", "age_0": "56", "age_1": "55", "age_3": "54"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = $225; priority = high.\n\nClaim to verify: \"the status is $225\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 17:00 to 22:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "5h00", "dur_3": "4h00", "dur_2": "5h15", "dur_0": "4h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 16:30 to 20:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "3h30", "dur_2": "3h30", "dur_0": "3h00", "dur_3": "2h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p1": "0.200", "p0": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = 2026-07-19; priority = high.\n\nClaim to verify: \"the owner is 2026-07-19\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.462", "p1": "0.462", "p0": "0.538", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=477, unit_price=477, discount_pct=0, region=EU, sku_total=954 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "227529", "v3": "228006", "v1": "954", "v2": "22752"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = $890; priority = high.\n\nClaim to verify: \"the amount is $890\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 134 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "62", "answer_0": "81", "answer_3": "65", "answer_1": "79"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1067 per year. It increases by 10% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "960.3", "answer_0": "1127.7", "answer_1": "1173.7", "answer_3": "1105.7"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 6 grams of salt per serving and serves 11 people. A shopper bought a 66 gram box and a 660 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = closed; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1969. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "57", "age_3": "54", "age_1": "55", "age_0": "56"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 227:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, April 17, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.222", "p0": "0.778", "p3": "0.750", "p1": "0.222"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 880:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $435. A fee of $176 is charged, then a deposit of $41 arrives. A second fee of $88 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "231", "answer_0": "212", "answer_3": "242", "answer_1": "206"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 133 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "132", "answer_2": "138", "answer_0": "120", "answer_1": "139"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = 2026-09-06; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = 2026-01-16; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $491 per year. It increases by 20% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "541.2", "answer_1": "589.2", "answer_3": "520.2", "answer_2": "392.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = 2026-04-28; priority = high.\n\nClaim to verify: \"the date is 2026-04-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $799. A fee of $162 is charged, then a deposit of $41 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "594", "answer_1": "584", "answer_0": "597", "answer_3": "617"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 271 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "7", "answer_0": "18", "answer_3": "7", "answer_2": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p1": "0.200", "p3": "0.050", "p2": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p2": "0.250", "p0": "0.333", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, October 18, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 15:15 to 16:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "1h15", "dur_2": "1h30", "dur_3": "0h30", "dur_1": "1h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = $191; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $692. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "35", "answer_0": "77", "answer_1": "41", "answer_3": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=841, unit_price=501, discount_pct=0, region=EU, sku_total=1342 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "42134", "v0": "421341", "v1": "1342", "v3": "422182"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 457:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "42", "answer_0": "84", "answer_3": "42", "answer_1": "48"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "20", "answer_0": "44", "answer_3": "24", "answer_1": "26"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $615. A fee of $166 is charged, then a deposit of $29 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "395", "answer_1": "366", "answer_3": "389", "answer_2": "371"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = 2026-07-26; priority = high.\n\nClaim to verify: \"the status is 2026-07-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, March 15, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1048 per year. It increases by 5% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1071.4", "answer_2": "995.6", "answer_1": "1100.4", "answer_0": "1059.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p3": "0.125", "p2": "0.250", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed February 20, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 26", "date_3": "February 28", "date_2": "February 28", "date_0": "March 02"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 06:45 to 19:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "12h45", "dur_0": "12h30", "dur_2": "12h45", "dur_3": "11h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = 2026-08-15; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Sunday, June 07, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 652:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = 2026-02-22; priority = high.\n\nClaim to verify: \"the date is 2026-02-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 08:45 to 14:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "5h15", "dur_1": "5h30", "dur_0": "4h45", "dur_3": "4h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 317 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "108", "answer_1": "118", "answer_2": "93", "answer_3": "123"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A recipe needs 7 grams of salt per serving and serves 6 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, May 20, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $344 per year. It increases by 20% this year, then a $49 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "363.8", "answer_2": "275.2", "answer_1": "412.8", "answer_3": "376.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = $472; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1032 per year. It increases by 5% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1030.6", "answer_1": "1083.6", "answer_2": "980.4", "answer_0": "1038.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $370 per year. It increases by 15% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "314.5", "answer_0": "409.5", "answer_3": "445.5", "answer_1": "425.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Contract signed September 06, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 11", "date_1": "September 09", "date_2": "September 11", "date_0": "September 09"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = 2026-07-05; priority = low.\n\nClaim to verify: \"the owner is 2026-07-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.250", "p2": "0.167", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = 2026-10-18; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 11:15 to 18:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "7h30", "dur_0": "7h00", "dur_3": "6h30", "dur_2": "7h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 261:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.533", "p2": "0.533", "p0": "0.467", "p3": "0.429"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 399 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "12", "answer_0": "32", "answer_2": "40", "answer_3": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed June 18, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 23", "date_2": "June 25", "date_1": "June 21", "date_0": "June 23"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1974. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "51", "age_2": "53", "age_3": "54", "age_0": "52"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $205 per year. It increases by 10% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "214.5", "answer_3": "253.5", "answer_2": "184.5", "answer_1": "225.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 372:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 3 grams of salt per serving and serves 10 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $721. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed March 26, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 03", "date_3": "April 01", "date_0": "April 01", "date_1": "March 30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 10 people. A shopper bought a 50 gram box and a 500 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = $657; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed February 06, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 22", "date_0": "February 20", "date_3": "February 18", "date_1": "February 16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 10 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $501. A fee of $147 is charged, then a deposit of $52 arrives. A second fee of $73 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "333", "answer_2": "318", "answer_3": "319", "answer_1": "330"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "42", "answer_3": "14", "answer_1": "44", "answer_0": "56"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed June 12, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 17", "date_3": "June 19", "date_2": "June 17", "date_0": "June 19"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed August 15, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 24", "date_2": "August 29", "date_3": "August 26", "date_0": "August 27"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $870. A fee of $151 is charged, then a deposit of $45 arrives. A second fee of $75 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "695", "answer_2": "681", "answer_1": "706", "answer_0": "689"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p1": "0.150", "p0": "0.100", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 4 grams of salt per serving and serves 7 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1846 per year. It increases by 25% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1384.5", "answer_1": "2307.5", "answer_0": "2282.5", "answer_3": "2277.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.750", "p2": "0.250", "p1": "0.250", "p3": "0.714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.364", "p1": "0.636", "p2": "0.636", "p3": "0.300"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = 2026-09-15; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.727", "p0": "0.273", "p1": "0.727", "p3": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $432. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Sunday, December 06, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p0": "0.429", "p2": "0.571", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 2 grams of salt per serving and serves 10 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 373 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "140", "answer_2": "138", "answer_3": "130", "answer_1": "136"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 955:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.083", "p1": "0.333", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = $97; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.250", "p0": "0.200", "p2": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1990. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "33", "age_0": "35", "age_1": "34", "age_2": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 273 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "336", "answer_2": "316", "answer_1": "342", "answer_0": "324"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, August 09, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "30", "answer_0": "72", "answer_3": "48", "answer_2": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed May 18, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "June 01", "date_2": "June 03", "date_3": "May 30", "date_1": "May 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 95 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_1": "124", "answer_2": "124", "answer_3": "108"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1974. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "50", "age_2": "53", "age_1": "51", "age_0": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.250", "p0": "0.333", "p1": "0.417"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 255 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "135", "answer_3": "122", "answer_2": "116", "answer_1": "150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = 2026-05-19; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=669, unit_price=556, discount_pct=0, region=EU, sku_total=1225 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "37196", "v0": "371964", "v3": "371295", "v1": "1225"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 252:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $370 per year. It increases by 10% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "333", "answer_0": "384", "answer_3": "397", "answer_1": "407"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1969. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "55", "age_2": "57", "age_0": "56", "age_3": "58"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.286", "p2": "0.714", "p1": "0.714"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1968. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "55", "age_2": "57", "age_0": "56", "age_3": "58"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1980. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "47", "age_0": "46", "age_1": "45", "age_3": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=662, unit_price=355, discount_pct=0, region=EU, sku_total=1017 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1017", "v2": "23501", "v3": "234348", "v0": "235010"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.100", "p1": "0.200", "p0": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1360 per year. It increases by 10% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1462", "answer_1": "1496", "answer_3": "1434", "answer_2": "1224"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = 2026-08-25; priority = low.\n\nClaim to verify: \"the owner is 2026-08-25\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $5k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $461. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1963. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "62", "age_3": "61", "age_2": "64", "age_0": "63"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 164 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "162", "answer_1": "178", "answer_2": "170", "answer_3": "152"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = 2026-11-25; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: status = 2026-03-07; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = 2026-11-24; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed February 18, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 22", "date_0": "February 24", "date_1": "February 22", "date_3": "February 24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = 2026-11-11; priority = high.\n\nClaim to verify: \"the date is 2026-11-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=402, unit_price=104, discount_pct=0, region=EU, sku_total=506 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "4180", "v1": "506", "v0": "41808", "v3": "41406"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, October 09, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 11:15 to 13:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "1h30", "dur_1": "2h30", "dur_0": "2h00", "dur_2": "2h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = 2026-09-20; priority = high.\n\nClaim to verify: \"the owner is 2026-09-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "23", "answer_0": "48", "answer_3": "30", "answer_2": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 3 grams of salt per serving and serves 9 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 920:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p0": "0.100", "p3": "0.050", "p2": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $258. A fee of $45 is charged, then a deposit of $58 arrives. A second fee of $22 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "249", "answer_3": "239", "answer_1": "232", "answer_2": "224"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=144, unit_price=538, discount_pct=0, region=EU, sku_total=682 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "77616", "v0": "77472", "v2": "7747", "v1": "682"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p2": "0.375", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 9 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $425. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $2k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=635, unit_price=512, discount_pct=0, region=EU, sku_total=1147 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "325120", "v3": "325755", "v1": "1147", "v2": "32512"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1963. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "62", "age_1": "60", "age_0": "61", "age_3": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 07:30 to 09:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h00", "dur_0": "1h45", "dur_2": "2h15", "dur_3": "1h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 337 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "154", "answer_0": "144", "answer_2": "124", "answer_3": "127"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1975. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "52", "age_0": "50", "age_2": "51", "age_1": "49"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed November 07, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 11", "date_0": "November 12", "date_2": "November 14", "date_3": "November 13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 694:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 407:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=969, unit_price=516, discount_pct=0, region=EU, sku_total=1485 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1485", "v2": "50000", "v3": "500973", "v0": "500004"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "40", "answer_1": "29", "answer_2": "24", "answer_0": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 9 grams of salt per serving and serves 7 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.250", "p3": "0.083", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1480 per year. It increases by 10% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1601", "answer_1": "1628", "answer_0": "1594", "answer_2": "1332"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = 2026-10-12; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1967. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "56", "age_0": "57", "age_3": "55", "age_2": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 320 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "12", "answer_3": "28", "answer_2": "7", "answer_1": "-8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $870. A fee of $197 is charged, then a deposit of $90 arrives. A second fee of $98 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "663", "answer_0": "665", "answer_3": "673", "answer_2": "651"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=540, unit_price=759, discount_pct=0, region=EU, sku_total=1299 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1299", "v0": "409860", "v2": "40986", "v3": "410400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1987. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "39", "age_0": "37", "age_1": "36", "age_2": "38"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.083", "p1": "0.417", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 583:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_3": "42", "answer_2": "30", "answer_1": "37"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 216 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "92", "answer_0": "72", "answer_3": "53", "answer_1": "84"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p3": "0.333", "p0": "0.400", "p2": "0.600"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=931, unit_price=473, discount_pct=0, region=EU, sku_total=1404 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "441294", "v1": "1404", "v2": "44036", "v0": "440363"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=353, unit_price=998, discount_pct=0, region=EU, sku_total=1351 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1351", "v3": "351941", "v2": "35229", "v0": "352294"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, April 09, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Monday, October 12, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Thursday, February 05, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $639. A fee of $41 is charged, then a deposit of $38 arrives. A second fee of $20 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "639", "answer_1": "604", "answer_3": "641", "answer_0": "616"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=690, unit_price=706, discount_pct=0, region=EU, sku_total=1396 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1396", "v2": "48714", "v0": "487140", "v3": "487830"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $27k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.250", "p2": "0.083", "p0": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 289:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $868 per year. It increases by 5% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "913.4", "answer_2": "824.6", "answer_1": "911.4", "answer_0": "877.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $796. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=188, unit_price=333, discount_pct=0, region=EU, sku_total=521 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "521", "v0": "62604", "v2": "6260", "v3": "62792"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.833", "p2": "0.500", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=788, unit_price=357, discount_pct=0, region=EU, sku_total=1145 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "28131", "v1": "1145", "v3": "280528", "v0": "281316"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 164 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "31", "answer_0": "36", "answer_2": "50", "answer_1": "42"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $563. A fee of $121 is charged, then a deposit of $53 arrives. A second fee of $60 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "410", "answer_1": "419", "answer_0": "435", "answer_3": "428"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 343:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $918 per year. It increases by 10% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "971.8", "answer_2": "826.2", "answer_3": "986.8", "answer_1": "1009.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_1": "19", "answer_0": "54", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $536. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed October 10, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 22", "date_0": "October 23", "date_2": "October 25", "date_1": "October 20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 260 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "240", "answer_2": "243", "answer_3": "255", "answer_1": "254"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.462", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $370. A fee of $52 is charged, then a deposit of $57 arrives. A second fee of $26 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "379", "answer_2": "365", "answer_0": "349", "answer_3": "346"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, November 03, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 292 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "216", "answer_3": "212", "answer_2": "236", "answer_1": "233"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $414. A fee of $113 is charged, then a deposit of $30 arrives. A second fee of $56 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "275", "answer_2": "264", "answer_3": "269", "answer_1": "283"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 4 grams of salt per serving and serves 3 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = 2026-06-16; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 363:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $519 per year. It increases by 20% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "622.8", "answer_0": "585.8", "answer_3": "601.8", "answer_2": "415.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed September 03, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 11", "date_0": "September 11", "date_2": "September 13", "date_1": "September 09"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $381 per year. It increases by 5% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "358.05", "answer_2": "361.95", "answer_1": "400.05", "answer_0": "388.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $251. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1974. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "50", "age_1": "49", "age_3": "52", "age_2": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $64. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $364. A fee of $141 is charged, then a deposit of $65 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "223", "answer_2": "241", "answer_3": "241", "answer_0": "218"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 169 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "139", "answer_0": "120", "answer_3": "107", "answer_2": "110"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed April 01, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 04", "date_0": "April 06", "date_2": "April 04", "date_3": "April 06"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 10:30 to 19:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "9h00", "dur_2": "9h00", "dur_3": "8h00", "dur_0": "8h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 5 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.167", "p3": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "15", "answer_0": "21", "answer_3": "9", "answer_2": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 8 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "26", "answer_3": "8", "answer_0": "32", "answer_2": "24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 766:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.778", "p0": "0.222", "p1": "0.778", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.150", "p0": "0.200", "p1": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = $415; priority = high.\n\nClaim to verify: \"the owner is $415\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.250", "p3": "0.083", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = $335; priority = high.\n\nClaim to verify: \"the date is $335\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1472 per year. It increases by 10% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1324.8", "answer_1": "1619.2", "answer_0": "1603.2", "answer_3": "1589.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = 2026-02-28; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.250", "p2": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 3 grams of salt per serving and serves 8 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 5 grams of salt per serving and serves 4 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $739 per year. It increases by 10% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "799.9", "answer_2": "665.1", "answer_1": "812.9", "answer_3": "776.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p3": "0.111", "p2": "0.800", "p1": "0.800"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed January 05, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 14", "date_0": "January 16", "date_2": "January 18", "date_3": "January 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=667, unit_price=497, discount_pct=0, region=EU, sku_total=1164 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1164", "v2": "33149", "v0": "331499", "v3": "332166"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = $863; priority = low.\n\nClaim to verify: \"the date is $863\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1985. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "42", "age_1": "40", "age_3": "39", "age_0": "41"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 890:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 392 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "20", "answer_3": "30", "answer_0": "32", "answer_2": "27"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $521. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $221. A fee of $192 is charged, then a deposit of $73 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "6", "answer_2": "-4", "answer_3": "15", "answer_1": "-21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed November 11, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 15", "date_2": "November 15", "date_3": "November 17", "date_0": "November 17"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.250", "p2": "0.167", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=825, unit_price=938, discount_pct=0, region=EU, sku_total=1763 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "77385", "v0": "773850", "v1": "1763", "v3": "774675"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1967. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_3": "60", "age_2": "59", "age_1": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $464. A fee of $93 is charged, then a deposit of $43 arrives. A second fee of $46 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "396", "answer_3": "388", "answer_2": "357", "answer_0": "368"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.625", "p3": "0.125", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $927 per year. It increases by 15% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1066.05", "answer_2": "787.95", "answer_3": "1060.05", "answer_0": "1042.05"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p1": "0.150", "p3": "0.050", "p2": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = $410; priority = low.\n\nClaim to verify: \"the status is $410\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=831, unit_price=933, discount_pct=0, region=EU, sku_total=1764 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "77532", "v1": "1764", "v0": "775323", "v3": "774492"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=638, unit_price=869, discount_pct=0, region=EU, sku_total=1507 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1507", "v3": "553784", "v0": "554422", "v2": "55442"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: status = ops; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 293 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "100", "answer_0": "108", "answer_1": "126", "answer_2": "99"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.429", "p1": "0.533", "p0": "0.467", "p2": "0.533"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = $882; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = 2026-02-02; priority = high.\n\nClaim to verify: \"the owner is 2026-02-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $779 per year. It increases by 10% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "701.1", "answer_3": "843.9", "answer_1": "856.9", "answer_0": "810.9"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.455", "p1": "0.545", "p2": "0.545", "p3": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 245:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed November 10, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 22", "date_1": "November 20", "date_2": "November 26", "date_0": "November 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 133 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "231", "answer_3": "216", "answer_1": "241", "answer_0": "224"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $805. A fee of $63 is charged, then a deposit of $87 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "775", "answer_0": "798", "answer_2": "808", "answer_3": "822"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = ops; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $613. A fee of $62 is charged, then a deposit of $42 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "558", "answer_3": "574", "answer_0": "562", "answer_2": "574"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = closed; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed February 06, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 19", "date_1": "February 15", "date_3": "February 17", "date_2": "February 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A shift runs from 15:15 to 22:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "6h00", "dur_1": "7h00", "dur_0": "6h15", "dur_2": "6h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 232 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "211", "answer_3": "196", "answer_0": "192", "answer_2": "173"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 3 grams of salt per serving and serves 6 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.444", "p1": "0.444", "p3": "0.500", "p0": "0.556"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 268 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_3": "62", "answer_2": "66", "answer_1": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 5 grams of salt per serving and serves 7 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 147 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "109", "answer_3": "115", "answer_0": "120", "answer_1": "124"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = 2026-09-10; priority = high.\n\nClaim to verify: \"the owner is 2026-09-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, November 21, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $260. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1487 per year. It increases by 10% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1587.7", "answer_1": "1635.7", "answer_2": "1338.3", "answer_3": "1582.7"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=752, unit_price=617, discount_pct=0, region=EU, sku_total=1369 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "46398", "v1": "1369", "v3": "464736", "v0": "463984"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 854:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $251. A fee of $88 is charged, then a deposit of $79 arrives. A second fee of $44 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "198", "answer_3": "195", "answer_1": "172", "answer_2": "204"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.083", "p1": "0.333", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $721 per year. It increases by 15% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "612.85", "answer_0": "788.15", "answer_1": "829.15", "answer_3": "757.15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 7 grams of salt per serving and serves 4 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = closed; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 17:15 to 20:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h45", "dur_3": "1h45", "dur_2": "2h30", "dur_0": "2h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 222:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = $525; priority = low.\n\nClaim to verify: \"the date is $525\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $485 per year. It increases by 5% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "509.25", "answer_2": "460.75", "answer_0": "473.25", "answer_3": "463.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 7 grams of salt per serving and serves 6 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed March 16, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 27", "date_2": "March 25", "date_1": "March 25", "date_0": "March 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 137 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "84", "answer_3": "107", "answer_0": "90", "answer_2": "81"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 83 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "140", "answer_2": "124", "answer_1": "152", "answer_3": "148"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "15", "answer_2": "30", "answer_0": "45", "answer_1": "33"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=180, unit_price=900, discount_pct=0, region=EU, sku_total=1080 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "162180", "v0": "162000", "v1": "1080", "v2": "16200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.556", "p3": "0.500", "p1": "0.444", "p2": "0.444"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed December 19, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 22", "date_0": "December 24", "date_1": "December 23", "date_3": "December 25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = open; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 396 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "288", "answer_3": "304", "answer_2": "301", "answer_1": "303"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $774 per year. It increases by 10% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "696.6", "answer_3": "811.4", "answer_0": "836.4", "answer_1": "851.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.167", "p1": "0.500", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 09:15 to 13:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "2h45", "dur_1": "3h45", "dur_2": "3h30", "dur_0": "3h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed April 13, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 22", "date_0": "April 22", "date_1": "April 20", "date_2": "April 24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, August 04, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 8 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 159 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "256", "answer_1": "237", "answer_2": "236", "answer_3": "249"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 333 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "88", "answer_2": "85", "answer_3": "89", "answer_0": "72"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed April 23, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 01", "date_3": "May 03", "date_0": "May 05", "date_2": "May 07"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=484, unit_price=596, discount_pct=0, region=EU, sku_total=1080 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "288464", "v1": "1080", "v3": "287980", "v2": "28846"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 3 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1722 per year. It increases by 25% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1291.5", "answer_1": "2152.5", "answer_0": "2128.5", "answer_3": "2117.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 981:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $321. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Thursday, August 06, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 12:00 to 16:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "4h15", "dur_3": "3h30", "dur_2": "4h30", "dur_1": "4h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 691:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed December 26, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 09", "date_3": "January 06", "date_0": "January 07", "date_1": "January 04"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 08:15 to 12:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "3h00", "dur_3": "2h45", "dur_1": "3h45", "dur_2": "3h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=645, unit_price=510, discount_pct=0, region=EU, sku_total=1155 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "32895", "v1": "1155", "v0": "328950", "v3": "328305"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=331, unit_price=403, discount_pct=0, region=EU, sku_total=734 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "133062", "v2": "13339", "v0": "133393", "v1": "734"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = $282; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 9 grams of salt per serving and serves 6 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 384 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "237", "answer_0": "252", "answer_3": "236", "answer_1": "255"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $594. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 7 grams of salt per serving and serves 10 people. A shopper bought a 70 gram box and a 700 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.533", "p0": "0.467", "p1": "0.533", "p3": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 429:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1977. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "51", "age_0": "49", "age_2": "50", "age_1": "48"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $330. A fee of $78 is charged, then a deposit of $55 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "264", "answer_0": "268", "answer_2": "286", "answer_1": "285"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $558. A fee of $141 is charged, then a deposit of $27 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "374", "answer_2": "394", "answer_1": "358", "answer_3": "395"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = $595; priority = high.\n\nClaim to verify: \"the date is $595\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 160 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "45", "answer_3": "53", "answer_1": "62", "answer_0": "48"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed February 11, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 26", "date_1": "February 20", "date_3": "February 22", "date_0": "February 24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=118, unit_price=164, discount_pct=0, region=EU, sku_total=282 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "1935", "v3": "19470", "v0": "19352", "v1": "282"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.615", "p0": "0.385", "p2": "0.615", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = 2026-12-01; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_0": "21", "answer_2": "9", "answer_1": "13"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = 2026-03-27; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.625", "p2": "0.375", "p1": "0.375", "p3": "0.571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=195, unit_price=217, discount_pct=0, region=EU, sku_total=412 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "42315", "v2": "4231", "v1": "412", "v3": "42510"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $708 per year. It increases by 20% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "824.6", "answer_2": "566.4", "answer_3": "864.6", "answer_1": "849.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = closed; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 971:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = $569; priority = high.\n\nClaim to verify: \"the amount is $569\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $37k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1142 per year. It increases by 15% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1276.3", "answer_3": "1294.3", "answer_2": "970.7", "answer_1": "1313.3"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 3 grams of salt per serving and serves 12 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed August 01, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 10", "date_1": "August 08", "date_2": "August 09", "date_0": "August 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 5 grams of salt per serving and serves 9 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed March 28, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 06", "date_0": "April 07", "date_1": "April 04", "date_2": "April 05"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1961. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "65", "age_1": "64", "age_3": "63", "age_2": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $592 per year. It increases by 10% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "606.2", "answer_3": "577.2", "answer_1": "651.2", "answer_2": "532.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $877. A fee of $161 is charged, then a deposit of $21 arrives. A second fee of $80 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "684", "answer_3": "672", "answer_1": "671", "answer_0": "657"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.625", "p0": "0.500", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = 2026-11-21; priority = low.\n\nClaim to verify: \"the status is 2026-11-21\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.444", "p0": "0.556", "p2": "0.444", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "9", "answer_0": "30", "answer_3": "21", "answer_1": "16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $612. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 111 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "69", "answer_0": "84", "answer_3": "97", "answer_1": "94"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=172, unit_price=646, discount_pct=0, region=EU, sku_total=818 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "111112", "v2": "11111", "v3": "111284", "v1": "818"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1980. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "45", "age_3": "48", "age_0": "46", "age_2": "47"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $427. A fee of $152 is charged, then a deposit of $74 arrives. A second fee of $76 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "278", "answer_1": "244", "answer_3": "303", "answer_0": "273"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1170 per year. It increases by 20% this year, then a $49 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "936", "answer_1": "1404", "answer_0": "1355", "answer_3": "1394"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 6 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=592, unit_price=789, discount_pct=0, region=EU, sku_total=1381 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "46708", "v0": "467088", "v3": "466496", "v1": "1381"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed November 05, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 18", "date_2": "November 20", "date_3": "November 16", "date_1": "November 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: date = 2026-07-02; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Wednesday, December 16, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $162. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = $502; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = $449; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed September 04, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 16", "date_1": "September 14", "date_0": "September 18", "date_2": "September 20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 5 grams of salt per serving and serves 12 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_0": "27", "answer_1": "15", "answer_2": "9"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1958. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "67", "age_3": "69", "age_2": "68", "age_1": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 536:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 228:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.571", "p0": "0.625", "p2": "0.375", "p1": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.533", "p0": "0.467", "p3": "0.429", "p2": "0.533"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $499. A fee of $142 is charged, then a deposit of $90 arrives. A second fee of $71 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "392", "answer_1": "399", "answer_0": "376", "answer_2": "389"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.375", "p3": "0.125", "p1": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $38k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Sunday, April 19, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 07:45 to 14:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "6h30", "dur_3": "5h45", "dur_1": "6h45", "dur_2": "6h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, March 17, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 71 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "76", "answer_3": "40", "answer_1": "46", "answer_0": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 13:45 to 19:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "4h15", "dur_0": "4h45", "dur_2": "5h00", "dur_1": "5h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1974. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "53", "age_0": "52", "age_1": "51", "age_3": "50"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 145 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "63", "answer_2": "84", "answer_3": "86", "answer_0": "80"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 248 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "74", "answer_3": "64", "answer_0": "60", "answer_2": "78"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 8 grams of salt per serving and serves 10 people. A shopper bought a 80 gram box and a 800 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=198, unit_price=999, discount_pct=0, region=EU, sku_total=1197 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "198000", "v1": "1197", "v2": "19780", "v0": "197802"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $466. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, June 04, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = $386; priority = low.\n\nClaim to verify: \"the amount is $386\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $528. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = 2026-01-17; priority = low.\n\nClaim to verify: \"the date is 2026-01-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.467", "p0": "0.500", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = 2026-09-07; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.538", "p3": "0.417", "p1": "0.538", "p0": "0.462"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 06:45 to 07:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "0h30", "dur_3": "-1h30", "dur_2": "0h15", "dur_0": "0h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = $417; priority = high.\n\nClaim to verify: \"the amount is $417\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: date = $103; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "15", "answer_3": "35", "answer_1": "22", "answer_0": "50"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $658. A fee of $144 is charged, then a deposit of $42 arrives. A second fee of $72 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "474", "answer_0": "484", "answer_2": "464", "answer_3": "511"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.455", "p1": "0.545", "p2": "0.545", "p3": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1756 per year. It increases by 20% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2076.2", "answer_3": "2100.2", "answer_1": "2107.2", "answer_2": "1404.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 340:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $159. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = $897; priority = high.\n\nClaim to verify: \"the amount is $897\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 183 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "258", "answer_1": "281", "answer_3": "268", "answer_0": "270"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1974. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "49", "age_2": "51", "age_3": "52", "age_0": "50"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = $607; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "66", "answer_2": "36", "answer_3": "30", "answer_1": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.538", "p1": "0.462", "p3": "0.500", "p2": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1962. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "65", "age_0": "64", "age_1": "63", "age_3": "66"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = $338; priority = high.\n\nClaim to verify: \"the date is $338\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = closed; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=671, unit_price=510, discount_pct=0, region=EU, sku_total=1181 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "341539", "v1": "1181", "v0": "342210", "v2": "34221"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=438, unit_price=826, discount_pct=0, region=EU, sku_total=1264 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "361350", "v0": "361788", "v1": "1264", "v2": "36178"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_1": "18", "answer_0": "36", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Contract signed May 16, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 20", "date_2": "May 23", "date_3": "May 22", "date_0": "May 21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p2": "0.375", "p3": "0.125", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 555:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=213, unit_price=283, discount_pct=0, region=EU, sku_total=496 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "6027", "v0": "60279", "v3": "60066", "v1": "496"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "6", "answer_1": "10", "answer_3": "12", "answer_0": "18"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=620, unit_price=904, discount_pct=0, region=EU, sku_total=1524 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "56048", "v0": "560480", "v1": "1524", "v3": "559860"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $396. A fee of $157 is charged, then a deposit of $41 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "175", "answer_2": "214", "answer_0": "202", "answer_3": "181"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 3 grams of salt per serving and serves 12 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed May 25, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 29", "date_3": "May 31", "date_0": "May 29", "date_2": "May 27"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p3": "0.167", "p2": "0.333", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1969. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "54", "age_2": "56", "age_3": "57", "age_0": "55"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 6 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = $561; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = 2026-09-16; priority = high.\n\nClaim to verify: \"the status is 2026-09-16\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed November 02, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 16", "date_3": "November 14", "date_1": "November 12", "date_2": "November 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 65 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "210", "answer_1": "204", "answer_2": "207", "answer_3": "208"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, May 24, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = 2026-08-02; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p1": "0.571", "p0": "0.429", "p2": "0.571"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 974:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1960. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "65", "age_0": "66", "age_2": "67", "age_3": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "35", "answer_2": "10", "answer_3": "25", "answer_1": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=153, unit_price=830, discount_pct=0, region=EU, sku_total=983 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "126837", "v1": "983", "v2": "12699", "v0": "126990"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1971. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "56", "age_3": "57", "age_0": "55", "age_1": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "10", "answer_3": "35", "answer_0": "45", "answer_1": "17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.750", "p0": "0.250", "p3": "0.143", "p2": "0.750"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_1": "31", "answer_0": "66", "answer_3": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "9", "answer_3": "21", "answer_1": "16", "answer_0": "30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = 2026-11-15; priority = high.\n\nClaim to verify: \"the amount is 2026-11-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=228, unit_price=141, discount_pct=0, region=EU, sku_total=369 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "32376", "v2": "3214", "v0": "32148", "v1": "369"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1971. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "54", "age_2": "56", "age_3": "53", "age_0": "55"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed March 17, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 25", "date_0": "March 27", "date_1": "March 25", "date_3": "March 27"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1964. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "61", "age_3": "62", "age_1": "59", "age_0": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.333", "p3": "0.167", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $785. A fee of $111 is charged, then a deposit of $46 arrives. A second fee of $55 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "674", "answer_0": "665", "answer_3": "671", "answer_2": "659"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, March 03, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 279 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "96", "answer_0": "80", "answer_1": "93", "answer_2": "86"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 328 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "176", "answer_1": "170", "answer_3": "191", "answer_0": "180"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed May 24, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 03", "date_0": "June 03", "date_2": "June 05", "date_1": "June 01"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 426:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $346. A fee of $192 is charged, then a deposit of $47 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "105", "answer_3": "94", "answer_2": "117", "answer_1": "107"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 387:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1573 per year. It increases by 15% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1778.95", "answer_1": "1808.95", "answer_0": "1795.95", "answer_2": "1337.05"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 19:45 to 22:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "3h00", "dur_3": "2h00", "dur_0": "2h45", "dur_2": "3h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 380:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = 2026-01-25; priority = low.\n\nClaim to verify: \"the owner is 2026-01-25\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = $702; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 7 people. A shopper bought a 14 gram box and a 140 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 342 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "44", "answer_3": "44", "answer_0": "54", "answer_2": "47"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=268, unit_price=747, discount_pct=0, region=EU, sku_total=1015 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "200464", "v0": "200196", "v1": "1015", "v2": "20019"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 828:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1976. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "48", "age_1": "47", "age_3": "50", "age_2": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 949:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = $380; priority = high.\n\nClaim to verify: \"the amount is $380\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1990. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "37", "age_1": "34", "age_0": "35", "age_2": "36"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1955 per year. It increases by 20% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2370", "answer_1": "2346", "answer_0": "2336", "answer_2": "1564"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, March 09, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1988. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "35", "age_3": "38", "age_0": "36", "age_2": "37"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = $924; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1983. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "40", "age_2": "42", "age_0": "41", "age_3": "39"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $19k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 5 grams of salt per serving and serves 6 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, May 19, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 347:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = 2026-05-13; priority = low.\n\nClaim to verify: \"the amount is 2026-05-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.150", "p3": "0.050", "p0": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, May 21, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p0": "0.500", "p3": "0.125", "p1": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An account starts at $701. A fee of $185 is charged, then a deposit of $61 arrives. A second fee of $92 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "489", "answer_1": "459", "answer_0": "485", "answer_2": "456"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1794 per year. It increases by 25% this year, then a $49 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2176.5", "answer_1": "2242.5", "answer_0": "2193.5", "answer_2": "1345.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $13k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 530:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=454, unit_price=872, discount_pct=0, region=EU, sku_total=1326 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "395888", "v3": "396342", "v2": "39588", "v1": "1326"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.400", "p1": "0.500", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $694 per year. It increases by 15% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "776.1", "answer_2": "589.9", "answer_3": "754.1", "answer_1": "798.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 394 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "285", "answer_2": "308", "answer_0": "288", "answer_3": "277"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.583", "p2": "0.385", "p0": "0.615", "p1": "0.385"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = $257; priority = low.\n\nClaim to verify: \"the date is $257\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = 2026-12-08; priority = high.\n\nClaim to verify: \"the amount is 2026-12-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=783, unit_price=736, discount_pct=0, region=EU, sku_total=1519 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "576288", "v1": "1519", "v2": "57628", "v3": "575505"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 16:00 to 18:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h00", "dur_3": "1h00", "dur_2": "1h30", "dur_0": "1h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1657 per year. It increases by 15% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1889.55", "answer_1": "1905.55", "answer_2": "1408.45", "answer_3": "1876.55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, November 13, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=689, unit_price=859, discount_pct=0, region=EU, sku_total=1548 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "591162", "v0": "591851", "v1": "1548", "v2": "59185"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, November 26, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 8 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 316 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "256", "answer_2": "262", "answer_3": "253", "answer_1": "239"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1966. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "58", "age_2": "61", "age_1": "59", "age_0": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "50", "answer_0": "64", "answer_2": "48", "answer_3": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 530:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 82 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "135", "answer_0": "140", "answer_2": "152", "answer_1": "129"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $224. A fee of $150 is charged, then a deposit of $65 arrives. A second fee of $75 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "91", "answer_0": "64", "answer_2": "66", "answer_3": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "42", "answer_1": "47", "answer_0": "77", "answer_3": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_2": "24", "answer_3": "48", "answer_1": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 6 grams of salt per serving and serves 9 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $502 per year. It increases by 5% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "494.1", "answer_2": "476.9", "answer_0": "514.1", "answer_1": "527.1"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 198 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "60", "answer_0": "72", "answer_1": "77", "answer_3": "55"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=845, unit_price=170, discount_pct=0, region=EU, sku_total=1015 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "144495", "v2": "14365", "v0": "143650", "v1": "1015"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 12:15 to 19:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "7h15", "dur_3": "6h15", "dur_2": "7h15", "dur_0": "6h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = 2026-08-01; priority = high.\n\nClaim to verify: \"the amount is 2026-08-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $83. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.333", "p0": "0.250", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = 2026-06-02; priority = high.\n\nClaim to verify: \"the amount is 2026-06-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.444", "p3": "0.500", "p2": "0.444", "p0": "0.556"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Tuesday, June 09, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = 2026-09-24; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.286", "p2": "0.286", "p0": "0.714", "p3": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.556", "p0": "0.444", "p1": "0.556", "p3": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=940, unit_price=687, discount_pct=0, region=EU, sku_total=1627 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "64578", "v1": "1627", "v0": "645780", "v3": "644840"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=839, unit_price=631, discount_pct=0, region=EU, sku_total=1470 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "529409", "v1": "1470", "v3": "530248", "v2": "52940"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = 2026-02-28; priority = high.\n\nClaim to verify: \"the amount is 2026-02-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = open; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: status = 2026-12-20; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1964. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "61", "age_1": "60", "age_2": "62", "age_3": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed February 24, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 25", "date_1": "February 27", "date_3": "March 01", "date_0": "February 27"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "20", "answer_2": "15", "answer_0": "40", "answer_3": "25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = $341; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=375, unit_price=741, discount_pct=0, region=EU, sku_total=1116 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "277875", "v3": "277500", "v2": "27787", "v1": "1116"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.150", "p0": "0.200", "p1": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "77", "answer_3": "35", "answer_2": "42", "answer_1": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $279. A fee of $150 is charged, then a deposit of $48 arrives. A second fee of $75 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "114", "answer_0": "102", "answer_2": "90", "answer_1": "109"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $486. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "28", "answer_0": "63", "answer_1": "33", "answer_3": "35"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = 2026-05-22; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.625", "p2": "0.333", "p0": "0.667", "p1": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1986. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "38", "age_3": "41", "age_0": "39", "age_2": "40"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Thursday, August 27, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $394 per year. It increases by 20% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "444.8", "answer_0": "437.8", "answer_2": "315.2", "answer_1": "472.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Contract signed March 02, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 08", "date_2": "March 12", "date_0": "March 10", "date_3": "March 10"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $39k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $2k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 09:00 to 10:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "1h30", "dur_3": "0h30", "dur_0": "1h00", "dur_2": "1h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 75 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "125", "answer_1": "142", "answer_3": "114", "answer_2": "110"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = 2026-05-13; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.222", "p3": "0.750", "p0": "0.778", "p1": "0.222"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 912:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = $144; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.429", "p2": "0.571", "p1": "0.571", "p3": "0.385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 990:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = $960; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $593. A fee of $192 is charged, then a deposit of $29 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "360", "answer_0": "334", "answer_2": "304", "answer_3": "320"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = $432; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = ops; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = 2026-06-12; priority = low.\n\nClaim to verify: \"the date is 2026-06-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Friday, February 13, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=168, unit_price=620, discount_pct=0, region=EU, sku_total=788 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "788", "v2": "10416", "v3": "104328", "v0": "104160"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1971. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "53", "age_2": "55", "age_3": "52", "age_0": "54"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 377 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "178", "answer_0": "189", "answer_3": "181", "answer_2": "198"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p1": "0.250", "p3": "0.083", "p2": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p0": "0.500", "p1": "0.625", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 8 grams of salt per serving and serves 10 people. A shopper bought a 80 gram box and a 800 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 11 people. A shopper bought a 33 gram box and a 330 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = $489; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.417", "p2": "0.250", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=633, unit_price=631, discount_pct=0, region=EU, sku_total=1264 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "398790", "v2": "39942", "v1": "1264", "v0": "399423"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1287 per year. It increases by 15% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1480.05", "answer_0": "1467.05", "answer_2": "1093.95", "answer_3": "1496.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.467", "p1": "0.500", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.385", "p2": "0.385", "p3": "0.583", "p0": "0.615"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.500", "p3": "0.167", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.200", "p3": "0.050", "p2": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, July 04, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=862, unit_price=546, discount_pct=0, region=EU, sku_total=1408 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "471514", "v0": "470652", "v2": "47065", "v1": "1408"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.500", "p2": "0.500", "p3": "0.467"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 382 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "222", "answer_2": "243", "answer_0": "225", "answer_3": "245"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = 2026-06-01; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: status = 2026-04-12; priority = high.\n\nClaim to verify: \"the status is 2026-04-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_0": "54", "answer_1": "19", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = $109; priority = high.\n\nClaim to verify: \"the amount is $109\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = 2026-04-05; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 661:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 295:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 891:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $744. A fee of $145 is charged, then a deposit of $39 arrives. A second fee of $72 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "566", "answer_2": "564", "answer_1": "548", "answer_3": "594"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.200", "p0": "0.150", "p2": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = ops; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 591:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = 2026-12-18; priority = low.\n\nClaim to verify: \"the amount is 2026-12-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = $144; priority = low.\n\nClaim to verify: \"the amount is $144\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = ops; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1797 per year. It increases by 5% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1707.15", "answer_0": "1872.85", "answer_3": "1885.85", "answer_1": "1886.85"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, November 02, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 315:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = $134; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 10:30 to 21:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "10h00", "dur_0": "10h45", "dur_2": "11h00", "dur_1": "11h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=825, unit_price=251, discount_pct=0, region=EU, sku_total=1076 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "207075", "v1": "1076", "v2": "20707", "v3": "207900"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = $940; priority = high.\n\nClaim to verify: \"the date is $940\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.250", "p2": "0.667", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed November 13, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 20", "date_2": "November 18", "date_3": "November 20", "date_1": "November 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An account starts at $588. A fee of $99 is charged, then a deposit of $21 arrives. A second fee of $49 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "458", "answer_3": "489", "answer_2": "438", "answer_0": "461"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=923, unit_price=615, discount_pct=0, region=EU, sku_total=1538 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1538", "v0": "567645", "v3": "566722", "v2": "56764"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, November 24, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p3": "0.200", "p1": "0.667", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $608. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $585. A fee of $132 is charged, then a deposit of $27 arrives. A second fee of $66 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "417", "answer_3": "435", "answer_1": "385", "answer_0": "414"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p3": "0.125", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.700", "p1": "0.300", "p2": "0.300", "p3": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p2": "0.250", "p0": "0.333", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $18k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p1": "0.500", "p3": "0.125", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "28", "answer_2": "25", "answer_3": "15", "answer_0": "40"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 12:00 to 16:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "4h15", "dur_2": "4h15", "dur_0": "4h00", "dur_3": "3h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = closed; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_1": "32", "answer_2": "25", "answer_0": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "21", "answer_1": "38", "answer_2": "35", "answer_0": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = 2026-01-19; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.400", "p1": "0.400", "p3": "0.556", "p0": "0.600"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed October 23, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 31", "date_2": "November 02", "date_3": "November 02", "date_0": "November 04"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A shift runs from 10:45 to 15:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "4h00", "dur_2": "4h15", "dur_1": "4h45", "dur_3": "3h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 6 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1968. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "60", "age_1": "57", "age_0": "58", "age_2": "59"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 999:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $866. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=826, unit_price=809, discount_pct=0, region=EU, sku_total=1635 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "668234", "v2": "66823", "v3": "667408", "v1": "1635"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed December 25, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 05", "date_0": "January 07", "date_1": "January 03", "date_2": "January 05"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=249, unit_price=931, discount_pct=0, region=EU, sku_total=1180 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "231819", "v2": "23181", "v3": "232068", "v1": "1180"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 88 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "200", "answer_0": "216", "answer_2": "220", "answer_3": "197"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $566. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.417", "p0": "0.333", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1742 per year. It increases by 5% this year, then a $40 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1829.1", "answer_3": "1755.1", "answer_2": "1654.9", "answer_0": "1789.1"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1985. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "39", "age_2": "40", "age_1": "38", "age_3": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: status = $325; priority = low.\n\nClaim to verify: \"the status is $325\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 3 grams of salt per serving and serves 10 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed December 23, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 02", "date_2": "January 04", "date_0": "January 06", "date_3": "January 04"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $449. A fee of $54 is charged, then a deposit of $30 arrives. A second fee of $27 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "398", "answer_3": "387", "answer_2": "395", "answer_1": "380"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $604. A fee of $196 is charged, then a deposit of $36 arrives. A second fee of $98 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "358", "answer_0": "346", "answer_3": "344", "answer_2": "337"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $233 per year. It increases by 15% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "198.05", "answer_1": "267.95", "answer_0": "253.95", "answer_3": "226.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 392 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "200", "answer_3": "202", "answer_1": "209", "answer_0": "189"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "49", "answer_3": "28", "answer_2": "21", "answer_1": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed January 15, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 26", "date_3": "January 24", "date_1": "January 22", "date_2": "January 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1390 per year. It increases by 15% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1598.5", "answer_0": "1585.5", "answer_2": "1181.5", "answer_3": "1574.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed November 03, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 10", "date_0": "November 10", "date_2": "November 12", "date_1": "November 08"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Wednesday, March 25, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 11:45 to 17:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "4h45", "dur_0": "4h30", "dur_1": "5h15", "dur_3": "4h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.615", "p0": "0.385", "p2": "0.615", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1981. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "42", "age_0": "43", "age_2": "44", "age_3": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 120 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "145", "answer_2": "148", "answer_3": "174", "answer_0": "162"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, March 14, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 229:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1962. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "66", "age_0": "64", "age_2": "65", "age_1": "63"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1771 per year. It increases by 25% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2213.75", "answer_3": "2223.75", "answer_0": "2199.75", "answer_2": "1328.25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $824. A fee of $166 is charged, then a deposit of $65 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "633", "answer_0": "640", "answer_3": "627", "answer_1": "657"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = $683; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.150", "p2": "0.100", "p1": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed July 02, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 11", "date_2": "July 13", "date_0": "July 15", "date_3": "July 13"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = 2026-03-04; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "20", "answer_1": "17", "answer_0": "32", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.167", "p1": "0.833", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 218 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "98", "answer_0": "100", "answer_2": "108", "answer_1": "117"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Thursday, November 19, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p1": "0.375", "p3": "0.125", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed December 12, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 19", "date_2": "December 20", "date_3": "December 21", "date_0": "December 22"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 67 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "56", "answer_2": "60", "answer_0": "54", "answer_3": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "43", "answer_3": "42", "answer_0": "78", "answer_2": "36"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = pending; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, January 28, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.125", "p2": "0.125", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.333", "p3": "0.636", "p1": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1982. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "43", "age_1": "41", "age_3": "40", "age_0": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 197 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "110", "answer_2": "88", "answer_0": "108", "answer_1": "116"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 478:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 349:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1991 per year. It increases by 5% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2090.55", "answer_0": "2068.55", "answer_2": "1891.45", "answer_3": "2091.55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 850:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.167", "p2": "0.083", "p1": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = $256; priority = high.\n\nClaim to verify: \"the amount is $256\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $828 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "786.6", "answer_3": "847.4", "answer_1": "869.4", "answer_0": "837.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 712:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p0": "0.167", "p1": "0.250", "p3": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_2": "12", "answer_3": "12", "answer_1": "14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.125", "p0": "0.375", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $584. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 832:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $8k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 3 grams of salt per serving and serves 10 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_2": "10", "answer_3": "30", "answer_1": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.364", "p2": "0.364", "p0": "0.636", "p3": "0.600"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=170, unit_price=597, discount_pct=0, region=EU, sku_total=767 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "767", "v3": "101320", "v0": "101490", "v2": "10149"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $630. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1969. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "57", "age_0": "56", "age_3": "58", "age_1": "55"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1965. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "62", "age_1": "60", "age_0": "61", "age_3": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = $656; priority = low.\n\nClaim to verify: \"the amount is $656\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_0": "30", "answer_3": "12", "answer_1": "20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p3": "0.429", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1592 per year. It increases by 25% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1953", "answer_1": "1990", "answer_2": "1194", "answer_0": "1967"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 345 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "69", "answer_0": "54", "answer_3": "57", "answer_1": "35"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed October 24, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 02", "date_1": "October 30", "date_2": "November 04", "date_3": "November 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "6", "answer_2": "18", "answer_1": "20", "answer_0": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = 2026-01-26; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 7 people. A shopper bought a 49 gram box and a 490 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, December 04, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.050", "p2": "0.150", "p0": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=259, unit_price=909, discount_pct=0, region=EU, sku_total=1168 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1168", "v3": "235690", "v2": "23543", "v0": "235431"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.150", "p0": "0.200", "p3": "0.050"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "48", "answer_1": "46", "answer_2": "40", "answer_0": "88"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $593. A fee of $155 is charged, then a deposit of $27 arrives. A second fee of $77 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "361", "answer_1": "367", "answer_3": "364", "answer_0": "388"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.333", "p0": "0.667", "p3": "0.636"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = 2026-07-07; priority = low.\n\nClaim to verify: \"the date is 2026-07-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=573, unit_price=375, discount_pct=0, region=EU, sku_total=948 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "21487", "v1": "948", "v0": "214875", "v3": "214302"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = pending; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $202. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $661. A fee of $62 is charged, then a deposit of $57 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "618", "answer_1": "633", "answer_0": "625", "answer_3": "611"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $638. A fee of $140 is charged, then a deposit of $61 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "473", "answer_0": "489", "answer_3": "510", "answer_2": "460"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 339:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 352:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, November 11, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $254. A fee of $114 is charged, then a deposit of $20 arrives. A second fee of $57 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "79", "answer_0": "103", "answer_1": "116", "answer_3": "99"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1533 per year. It increases by 25% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1149.75", "answer_1": "1916.25", "answer_3": "1881.25", "answer_0": "1891.25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = open; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, August 20, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = $288; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1177 per year. It increases by 5% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1222.85", "answer_2": "1118.15", "answer_1": "1235.85", "answer_3": "1249.85"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 4 people. A shopper bought a 8 gram box and a 80 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1962. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "63", "age_3": "66", "age_0": "64", "age_2": "65"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p0": "0.500", "p1": "0.625", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed February 11, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 20", "date_0": "February 18", "date_1": "February 16", "date_3": "February 18"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=576, unit_price=859, discount_pct=0, region=EU, sku_total=1435 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "494208", "v0": "494784", "v1": "1435", "v2": "49478"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = $45; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed January 03, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 11", "date_0": "January 14", "date_2": "January 16", "date_3": "January 13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed December 09, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 20", "date_1": "December 18", "date_0": "December 22", "date_2": "December 24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed May 24, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 29", "date_1": "May 29", "date_2": "May 27", "date_3": "May 31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $694. A fee of $190 is charged, then a deposit of $62 arrives. A second fee of $95 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "451", "answer_1": "490", "answer_2": "499", "answer_0": "471"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=653, unit_price=384, discount_pct=0, region=EU, sku_total=1037 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "250099", "v0": "250752", "v1": "1037", "v2": "25075"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=816, unit_price=262, discount_pct=0, region=EU, sku_total=1078 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "212976", "v0": "213792", "v2": "21379", "v1": "1078"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.400", "p1": "0.400", "p0": "0.600", "p3": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = ops; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = $907; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 7 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $827. A fee of $163 is charged, then a deposit of $90 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "647", "answer_2": "645", "answer_3": "643", "answer_0": "673"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = open; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.500", "p0": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1725 per year. It increases by 10% this year, then a $40 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1863.5", "answer_0": "1857.5", "answer_2": "1552.5", "answer_1": "1897.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 205:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 06:15 to 22:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "15h45", "dur_1": "15h45", "dur_0": "15h15", "dur_3": "14h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 886:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p0": "0.150", "p1": "0.200", "p3": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 11:30 to 15:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "3h45", "dur_0": "3h00", "dur_3": "2h45", "dur_2": "3h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $767. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = 2026-03-19; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.083", "p0": "0.333", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 6 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 9 grams of salt per serving and serves 8 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1960. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "68", "age_0": "66", "age_2": "67", "age_1": "65"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.667", "p3": "0.167", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $557. A fee of $51 is charged, then a deposit of $74 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "555", "answer_3": "579", "answer_1": "558", "answer_2": "560"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = 2026-05-02; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "21", "answer_3": "42", "answer_0": "63", "answer_1": "27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = 2026-04-10; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=361, unit_price=471, discount_pct=0, region=EU, sku_total=832 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "17003", "v0": "170031", "v3": "170392", "v1": "832"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.571", "p1": "0.429", "p2": "0.429", "p3": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=157, unit_price=809, discount_pct=0, region=EU, sku_total=966 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "12701", "v0": "127013", "v1": "966", "v3": "127170"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $444. A fee of $198 is charged, then a deposit of $73 arrives. A second fee of $99 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "220", "answer_1": "207", "answer_3": "228", "answer_2": "237"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 6 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.200", "p1": "0.727", "p0": "0.273", "p2": "0.727"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 3 grams of salt per serving and serves 6 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=926, unit_price=673, discount_pct=0, region=EU, sku_total=1599 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "62319", "v3": "622272", "v1": "1599", "v0": "623198"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed December 09, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 14", "date_0": "December 16", "date_2": "December 18", "date_3": "December 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = 2026-12-19; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 102 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "236", "answer_2": "227", "answer_0": "240", "answer_1": "221"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "27", "answer_2": "15", "answer_1": "19", "answer_3": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 10 people. A shopper bought a 50 gram box and a 500 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = 2026-08-22; priority = low.\n\nClaim to verify: \"the date is 2026-08-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1975. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "53", "age_0": "51", "age_1": "50", "age_2": "52"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_2": "20", "answer_1": "24", "answer_3": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, May 10, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Saturday, July 11, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 638:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = 2026-12-28; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1698 per year. It increases by 25% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2122.5", "answer_3": "2064.5", "answer_2": "1273.5", "answer_0": "2092.5"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p3": "0.050", "p1": "0.250", "p2": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed December 08, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 16", "date_3": "December 16", "date_1": "December 14", "date_2": "December 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_2": "24", "answer_1": "26", "answer_0": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed April 19, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 26", "date_3": "April 26", "date_0": "April 24", "date_1": "April 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.333", "p3": "0.083", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Saturday, November 14, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $4k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p0": "0.333", "p3": "0.083", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = $605; priority = low.\n\nClaim to verify: \"the amount is $605\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 5 grams of salt per serving and serves 4 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $485. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $475. A fee of $54 is charged, then a deposit of $20 arrives. A second fee of $27 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "406", "answer_2": "443", "answer_0": "414", "answer_3": "403"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $3k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 802:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.333", "p1": "0.333", "p3": "0.636"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, February 21, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.667", "p2": "0.667", "p3": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 82 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "252", "answer_2": "235", "answer_1": "266", "answer_3": "241"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $32k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = $308; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $222. A fee of $187 is charged, then a deposit of $86 arrives. A second fee of $93 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1", "answer_3": "11", "answer_0": "28", "answer_1": "22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = 2026-12-10; priority = high.\n\nClaim to verify: \"the date is 2026-12-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "32", "answer_2": "24", "answer_1": "28", "answer_0": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "70", "answer_3": "28", "answer_2": "42", "answer_1": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Thursday, June 04, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed July 26, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 02", "date_2": "July 29", "date_0": "July 31", "date_1": "July 31"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "20", "answer_0": "56", "answer_2": "14", "answer_3": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 7 grams of salt per serving and serves 10 people. A shopper bought a 70 gram box and a 700 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, September 05, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1966. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "60", "age_1": "59", "age_3": "62", "age_2": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $452. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.636", "p0": "0.667", "p2": "0.333", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Wednesday, November 25, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 368 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "212", "answer_0": "196", "answer_3": "185", "answer_2": "205"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $545. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, May 09, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, September 20, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $32k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.600", "p2": "0.400", "p3": "0.556", "p1": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1366 per year. It increases by 25% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1024.5", "answer_0": "1689.5", "answer_3": "1651.5", "answer_1": "1707.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $763 per year. It increases by 15% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "648.55", "answer_0": "840.45", "answer_1": "877.45", "answer_3": "865.45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $34k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "10", "answer_0": "35", "answer_2": "25", "answer_1": "27"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 370 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "180", "answer_0": "175", "answer_1": "188", "answer_3": "177"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed June 14, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 24", "date_0": "June 26", "date_2": "June 28", "date_3": "June 26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "27", "answer_1": "21", "answer_3": "9", "answer_2": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = 2026-11-13; priority = low.\n\nClaim to verify: \"the amount is 2026-11-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $35k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=543, unit_price=614, discount_pct=0, region=EU, sku_total=1157 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "333402", "v1": "1157", "v2": "33340", "v3": "333945"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=631, unit_price=274, discount_pct=0, region=EU, sku_total=905 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "172894", "v1": "905", "v2": "17289", "v3": "173525"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = 2026-12-19; priority = high.\n\nClaim to verify: \"the status is 2026-12-19\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed August 21, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "August 29", "date_3": "August 27", "date_1": "August 25", "date_0": "August 27"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 253 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_2": "47", "answer_3": "61", "answer_1": "37"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 713:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1804 per year. It increases by 5% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1894.2", "answer_0": "1880.2", "answer_2": "1713.8", "answer_3": "1851.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.083", "p3": "0.083", "p0": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 4 grams of salt per serving and serves 11 people. A shopper bought a 44 gram box and a 440 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = pending; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1980. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "45", "age_3": "47", "age_1": "44", "age_2": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = $845; priority = high.\n\nClaim to verify: \"the amount is $845\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $378 per year. It increases by 10% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "340.2", "answer_0": "397.8", "answer_3": "431.8", "answer_1": "415.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1985. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "39", "age_3": "42", "age_2": "41", "age_0": "40"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 09:15 to 10:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "1h30", "dur_2": "1h45", "dur_3": "0h30", "dur_0": "1h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1964. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "60", "age_1": "61", "age_0": "62", "age_2": "63"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed October 06, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "October 19", "date_1": "October 15", "date_3": "October 17", "date_2": "October 17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=922, unit_price=734, discount_pct=0, region=EU, sku_total=1656 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1656", "v3": "675826", "v2": "67674", "v0": "676748"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1394 per year. It increases by 10% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1533.4", "answer_0": "1519.4", "answer_3": "1500.4", "answer_2": "1254.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, May 09, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.333", "p2": "0.167", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1958. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "66", "age_2": "68", "age_0": "67", "age_3": "69"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p3": "0.625", "p1": "0.333", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.636", "p2": "0.636", "p3": "0.300", "p0": "0.364"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 442:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=474, unit_price=356, discount_pct=0, region=EU, sku_total=830 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "16874", "v0": "168744", "v3": "168270", "v1": "830"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = $229; priority = high.\n\nClaim to verify: \"the owner is $229\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 4 grams of salt per serving and serves 9 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = 2026-05-05; priority = high.\n\nClaim to verify: \"the amount is 2026-05-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 249 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "8", "answer_3": "33", "answer_0": "18", "answer_2": "13"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.467", "p2": "0.500", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=778, unit_price=432, discount_pct=0, region=EU, sku_total=1210 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "336096", "v1": "1210", "v2": "33609", "v3": "335318"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1927 per year. It increases by 15% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2165.05", "answer_0": "2180.05", "answer_1": "2216.05", "answer_2": "1637.95"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p2": "0.150", "p3": "0.050", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed January 05, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 10", "date_0": "January 08", "date_2": "January 10", "date_1": "January 08"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $115. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 644:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "10", "answer_2": "30", "answer_1": "32", "answer_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Thursday, May 07, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Wednesday, June 03, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 12 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 71 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "142", "answer_1": "124", "answer_0": "144", "answer_2": "164"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $572. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed July 08, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 17", "date_0": "July 15", "date_1": "July 13", "date_3": "July 15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.429", "p3": "0.333", "p1": "0.571", "p2": "0.571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = 2026-05-06; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p2": "0.571", "p3": "0.333", "p0": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 389 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "108", "answer_2": "118", "answer_1": "130", "answer_0": "120"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "28", "answer_3": "49", "answer_0": "70", "answer_2": "21"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $472. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = $786; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 938:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, June 22, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=130, unit_price=299, discount_pct=0, region=EU, sku_total=429 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "38870", "v2": "3887", "v1": "429", "v3": "39000"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=196, unit_price=780, discount_pct=0, region=EU, sku_total=976 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "15288", "v1": "976", "v3": "153076", "v0": "152880"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed October 16, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 25", "date_1": "October 23", "date_3": "October 25", "date_0": "October 27"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=677, unit_price=158, discount_pct=0, region=EU, sku_total=835 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "106966", "v1": "835", "v3": "106289", "v2": "10696"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $203. A fee of $122 is charged, then a deposit of $33 arrives. A second fee of $61 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_1": "62", "answer_0": "53", "answer_3": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1156 per year. It increases by 15% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1264.4", "answer_2": "982.6", "answer_1": "1329.4", "answer_0": "1279.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "16", "answer_1": "10", "answer_3": "8", "answer_2": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 7 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1986. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "39", "age_1": "37", "age_0": "38", "age_3": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = 2026-11-25; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $14k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $289. A fee of $147 is charged, then a deposit of $41 arrives. A second fee of $73 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "100", "answer_1": "137", "answer_0": "110", "answer_3": "94"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, June 01, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed February 22, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 03", "date_3": "March 05", "date_2": "March 03", "date_0": "March 05"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: date = $177; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p2": "0.375", "p0": "0.500", "p3": "0.125"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 280 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "200", "answer_0": "210", "answer_3": "206", "answer_2": "203"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "25", "answer_2": "20", "answer_1": "25", "answer_0": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $4k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_1": "16", "answer_3": "30", "answer_2": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 224 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "68", "answer_0": "72", "answer_3": "57", "answer_1": "58"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1989. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "37", "age_3": "34", "age_1": "35", "age_0": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = 2026-06-05; priority = high.\n\nClaim to verify: \"the amount is 2026-06-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 873:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.273", "p3": "0.700", "p0": "0.727", "p2": "0.273"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed January 22, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 27", "date_3": "January 29", "date_0": "January 29", "date_1": "January 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $807. A fee of $68 is charged, then a deposit of $68 arrives. A second fee of $34 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "773", "answer_1": "778", "answer_3": "799", "answer_2": "743"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p0": "0.500", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_3": "20", "answer_1": "29", "answer_0": "44"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 6 grams of salt per serving and serves 7 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, July 11, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1983. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "40", "age_2": "42", "age_3": "43", "age_0": "41"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=374, unit_price=719, discount_pct=0, region=EU, sku_total=1093 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "26890", "v3": "268532", "v1": "1093", "v0": "268906"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = $616; priority = low.\n\nClaim to verify: \"the amount is $616\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=643, unit_price=537, discount_pct=0, region=EU, sku_total=1180 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "345934", "v0": "345291", "v1": "1180", "v2": "34529"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.375", "p1": "0.625", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.429", "p2": "0.571", "p3": "0.385", "p1": "0.571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1228 per year. It increases by 15% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1043.8", "answer_0": "1396.2", "answer_1": "1412.2", "answer_3": "1363.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 4 grams of salt per serving and serves 7 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "8", "answer_0": "12", "answer_2": "6", "answer_3": "6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 15:00 to 19:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "4h45", "dur_0": "4h15", "dur_2": "4h30", "dur_3": "3h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "14", "answer_1": "19", "answer_0": "49", "answer_3": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = 2026-11-24; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 11 people. A shopper bought a 88 gram box and a 880 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p0": "0.500", "p3": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p3": "0.111", "p1": "0.800", "p2": "0.800"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 533:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 826:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.333", "p0": "0.500", "p1": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 255 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "67", "answer_2": "59", "answer_3": "67", "answer_0": "72"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = 2026-06-24; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 06:45 to 14:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "7h30", "dur_1": "8h00", "dur_2": "8h00", "dur_3": "7h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1384 per year. It increases by 20% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1107.2", "answer_0": "1632.8", "answer_3": "1653.8", "answer_1": "1660.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $381. A fee of $156 is charged, then a deposit of $25 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "172", "answer_3": "160", "answer_2": "187", "answer_1": "188"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed June 17, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 23", "date_2": "June 23", "date_3": "June 25", "date_0": "June 25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, September 26, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = $121; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Thursday, October 01, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1983. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "42", "age_3": "40", "age_1": "41", "age_2": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p0": "0.500", "p3": "0.125", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 735:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1977. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "48", "age_2": "49", "age_1": "47", "age_3": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 310 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "89", "answer_0": "70", "answer_2": "56", "answer_3": "73"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = 2026-08-22; priority = high.\n\nClaim to verify: \"the owner is 2026-08-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.462", "p2": "0.538", "p3": "0.417", "p1": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.538", "p3": "0.417", "p2": "0.538", "p0": "0.462"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $809. A fee of $101 is charged, then a deposit of $57 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "713", "answer_0": "715", "answer_3": "700", "answer_2": "737"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $8k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 6 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 6 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "32", "answer_0": "60", "answer_3": "35", "answer_2": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $470. A fee of $49 is charged, then a deposit of $28 arrives. A second fee of $24 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "401", "answer_2": "406", "answer_3": "406", "answer_0": "425"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1978. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "48", "age_2": "49", "age_1": "47", "age_3": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 367 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "66", "answer_3": "88", "answer_0": "72", "answer_1": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed August 17, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 29", "date_0": "August 31", "date_2": "September 02", "date_1": "August 27"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $463. A fee of $85 is charged, then a deposit of $46 arrives. A second fee of $42 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "379", "answer_2": "365", "answer_3": "391", "answer_0": "382"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = 2026-12-02; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $589 per year. It increases by 20% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "471.2", "answer_3": "713.8", "answer_1": "706.8", "answer_0": "687.8"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = $907; priority = high.\n\nClaim to verify: \"the status is $907\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $109; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=189, unit_price=397, discount_pct=0, region=EU, sku_total=586 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "7503", "v3": "75222", "v0": "75033", "v1": "586"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, May 07, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $448. A fee of $55 is charged, then a deposit of $23 arrives. A second fee of $27 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "407", "answer_2": "376", "answer_0": "389", "answer_3": "359"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.714", "p3": "0.667", "p1": "0.286", "p2": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1458 per year. It increases by 25% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1822.5", "answer_0": "1796.5", "answer_2": "1093.5", "answer_3": "1833.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.636", "p0": "0.364", "p1": "0.636", "p3": "0.300"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1981. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "45", "age_1": "44", "age_3": "43", "age_2": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 7 grams of salt per serving and serves 5 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = pending; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 243:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 9 grams of salt per serving and serves 10 people. A shopper bought a 90 gram box and a 900 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = $69; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "42", "answer_1": "46", "answer_0": "70", "answer_3": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.333", "p3": "0.083", "p1": "0.417"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed February 17, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 23", "date_1": "February 21", "date_2": "February 21", "date_3": "February 23"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = $735; priority = high.\n\nClaim to verify: \"the owner is $735\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, July 14, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1959. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "67", "age_0": "65", "age_2": "66", "age_1": "64"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $448. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed March 15, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 20", "date_1": "March 20", "date_3": "March 22", "date_2": "March 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "65", "age_1": "63", "age_0": "64", "age_3": "62"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1987. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "35", "age_2": "38", "age_0": "37", "age_1": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1884 per year. It increases by 20% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1507.2", "answer_0": "2232.8", "answer_1": "2260.8", "answer_3": "2238.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_2": "28", "answer_1": "34", "answer_0": "70"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p1": "0.333", "p2": "0.333", "p3": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.250", "p3": "0.083", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 364:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = $230; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = 2026-06-07; priority = high.\n\nClaim to verify: \"the status is 2026-06-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p3": "0.050", "p0": "0.200", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 75 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "420", "answer_0": "432", "answer_2": "445", "answer_1": "417"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 15:30 to 18:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h00", "dur_3": "2h15", "dur_0": "2h45", "dur_1": "3h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $200. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 70 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "383", "answer_3": "365", "answer_0": "378", "answer_1": "395"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 750:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "14", "answer_0": "32", "answer_2": "8", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.500", "p3": "0.467", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p2": "0.250", "p1": "0.500", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 332:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 06:45 to 13:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "6h30", "dur_0": "6h00", "dur_1": "6h15", "dur_3": "5h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = $516; priority = low.\n\nClaim to verify: \"the status is $516\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1961. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "64", "age_1": "62", "age_3": "61", "age_0": "63"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.250", "p1": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = 2026-06-18; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1890 per year. It increases by 10% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2079", "answer_0": "2041", "answer_3": "2062", "answer_2": "1701"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = $685; priority = high.\n\nClaim to verify: \"the date is $685\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.286", "p3": "0.667", "p0": "0.714", "p1": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $458. A fee of $115 is charged, then a deposit of $55 arrives. A second fee of $57 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "322", "answer_2": "333", "answer_1": "358", "answer_0": "341"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 13:00 to 22:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "9h00", "dur_1": "9h30", "dur_3": "8h30", "dur_2": "9h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p2": "0.600", "p3": "0.333", "p1": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.429", "p0": "0.571", "p3": "0.500", "p1": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Wednesday, March 25, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 7 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = 2026-01-26; priority = low.\n\nClaim to verify: \"the date is 2026-01-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $684. A fee of $130 is charged, then a deposit of $66 arrives. A second fee of $65 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "583", "answer_2": "573", "answer_1": "529", "answer_0": "555"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, May 09, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.150", "p2": "0.100", "p1": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.417", "p3": "0.083", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p3": "0.444", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, December 25, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.750", "p1": "0.750", "p0": "0.250", "p3": "0.143"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $870 per year. It increases by 10% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "783", "answer_0": "922", "answer_3": "952", "answer_1": "957"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1989. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "33", "age_0": "35", "age_2": "36", "age_1": "34"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed November 03, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 15", "date_1": "November 11", "date_3": "November 13", "date_0": "November 13"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.250", "p1": "0.667", "p2": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed November 16, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 25", "date_0": "November 27", "date_2": "November 25", "date_3": "November 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p1": "0.200", "p3": "0.050", "p0": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = $845; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.167", "p1": "0.833", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.467", "p3": "0.500", "p1": "0.467", "p0": "0.533"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p0": "0.100", "p2": "0.050", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 103 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_1": "97", "answer_2": "69", "answer_3": "95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = 2026-02-01; priority = low.\n\nClaim to verify: \"the amount is 2026-02-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.083", "p3": "0.083", "p0": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed July 23, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "August 04", "date_1": "July 31", "date_3": "August 02", "date_2": "August 06"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.462", "p1": "0.538", "p2": "0.538", "p3": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1979. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "45", "age_1": "46", "age_0": "47", "age_2": "48"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $16k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed July 17, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 21", "date_3": "July 23", "date_2": "July 21", "date_0": "July 23"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 7 grams of salt per serving and serves 9 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "45", "answer_2": "40", "answer_3": "40", "answer_0": "80"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.778", "p0": "0.222", "p3": "0.125", "p2": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = 2026-10-09; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $560. A fee of $185 is charged, then a deposit of $90 arrives. A second fee of $92 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "371", "answer_0": "373", "answer_1": "362", "answer_2": "356"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $19k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.200", "p0": "0.150", "p2": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p0": "0.429", "p2": "0.571", "p3": "0.385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_3": "18", "answer_1": "12", "answer_2": "6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 8 grams of salt per serving and serves 6 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.467", "p1": "0.500", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=364, unit_price=787, discount_pct=0, region=EU, sku_total=1151 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "28646", "v3": "286104", "v1": "1151", "v0": "286468"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1088 per year. It increases by 10% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1150.8", "answer_0": "1158.8", "answer_2": "979.2", "answer_1": "1196.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = 2026-04-06; priority = low.\n\nClaim to verify: \"the date is 2026-04-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1931 per year. It increases by 10% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2089.1", "answer_1": "2124.1", "answer_2": "1737.9", "answer_3": "2107.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.333", "p1": "0.500", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1959. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "66", "age_2": "68", "age_3": "65", "age_0": "67"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 152 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "75", "answer_1": "58", "answer_0": "70", "answer_3": "88"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = open; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = 2026-10-28; priority = low.\n\nClaim to verify: \"the status is 2026-10-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = $327; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed August 08, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "August 13", "date_3": "August 14", "date_1": "August 12", "date_2": "August 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 342 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "81", "answer_3": "64", "answer_1": "88", "answer_0": "72"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 860:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: owner = $652; priority = high.\n\nClaim to verify: \"the owner is $652\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 202:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, October 04, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = $117; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "91", "answer_2": "42", "answer_3": "49", "answer_1": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1776 per year. It increases by 5% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1864.8", "answer_2": "1687.2", "answer_3": "1857.8", "answer_0": "1826.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed November 02, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 06", "date_0": "November 06", "date_3": "November 08", "date_2": "November 04"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $395. A fee of $160 is charged, then a deposit of $49 arrives. A second fee of $80 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "204", "answer_2": "219", "answer_1": "192", "answer_3": "191"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $580. A fee of $127 is charged, then a deposit of $85 arrives. A second fee of $63 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "458", "answer_3": "448", "answer_0": "475", "answer_1": "486"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 272:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed November 12, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 24", "date_2": "November 28", "date_1": "November 22", "date_0": "November 26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Thursday, June 04, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.615", "p1": "0.615", "p0": "0.385", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed June 27, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 05", "date_2": "July 08", "date_0": "July 06", "date_1": "July 03"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = pending; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.375", "p3": "0.125", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = 2026-09-04; priority = low.\n\nClaim to verify: \"the amount is 2026-09-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed January 20, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 03", "date_1": "January 30", "date_2": "February 01", "date_3": "February 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $836. A fee of $65 is charged, then a deposit of $75 arrives. A second fee of $32 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "825", "answer_3": "794", "answer_1": "819", "answer_0": "814"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1974. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "51", "age_0": "52", "age_3": "54", "age_2": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=839, unit_price=414, discount_pct=0, region=EU, sku_total=1253 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "347346", "v1": "1253", "v2": "34734", "v3": "346507"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 496:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.600", "p2": "0.400", "p1": "0.400", "p3": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1970. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "57", "age_3": "54", "age_0": "56", "age_1": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 363 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "39", "answer_3": "38", "answer_1": "42", "answer_0": "50"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1981. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "46", "age_1": "43", "age_2": "45", "age_0": "44"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = 2026-11-17; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=680, unit_price=404, discount_pct=0, region=EU, sku_total=1084 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "274040", "v0": "274720", "v2": "27472", "v1": "1084"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.500", "p1": "0.667", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed August 15, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 26", "date_2": "August 29", "date_1": "August 24", "date_0": "August 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 7 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 14, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 24", "date_1": "April 22", "date_0": "April 24", "date_2": "April 22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed January 27, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 03", "date_2": "February 03", "date_3": "February 05", "date_0": "February 05"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: status = 2026-05-06; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 691:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = $560; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 4 grams of salt per serving and serves 9 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.167", "p1": "0.333", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed October 17, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 20", "date_1": "October 21", "date_3": "October 23", "date_0": "October 22"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $653 per year. It increases by 20% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "522.4", "answer_3": "716.6", "answer_0": "756.6", "answer_1": "783.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 943:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = open; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Sunday, November 22, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $405. A fee of $103 is charged, then a deposit of $76 arrives. A second fee of $51 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "299", "answer_0": "327", "answer_3": "300", "answer_2": "334"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $40k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed July 24, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "August 07", "date_0": "August 05", "date_1": "August 01", "date_3": "August 03"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $601. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $228. A fee of $156 is charged, then a deposit of $75 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "61", "answer_0": "69", "answer_3": "45", "answer_1": "76"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 4 grams of salt per serving and serves 9 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1974. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "52", "age_3": "54", "age_2": "53", "age_1": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.545", "p0": "0.583", "p2": "0.417", "p1": "0.417"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "40", "answer_0": "72", "answer_3": "32", "answer_1": "44"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1163 per year. It increases by 25% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1424.75", "answer_2": "872.25", "answer_1": "1453.75", "answer_3": "1451.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 96 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "116", "answer_1": "140", "answer_3": "134", "answer_0": "120"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Friday, December 18, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 932:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = 2026-04-25; priority = high.\n\nClaim to verify: \"the date is 2026-04-25\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = 2026-04-25; priority = high.\n\nClaim to verify: \"the status is 2026-04-25\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $18k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 427:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "24", "answer_0": "36", "answer_2": "18", "answer_3": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $665. A fee of $197 is charged, then a deposit of $57 arrives. A second fee of $98 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "455", "answer_3": "438", "answer_0": "427", "answer_2": "413"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 231 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "205", "answer_1": "203", "answer_3": "203", "answer_0": "192"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = $448; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed September 04, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 20", "date_3": "September 16", "date_1": "September 14", "date_0": "September 18"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 09:30 to 12:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h45", "dur_1": "3h15", "dur_3": "2h15", "dur_2": "3h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Saturday, April 11, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "36", "answer_2": "36", "answer_0": "72", "answer_1": "42"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1977. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "46", "age_2": "49", "age_0": "48", "age_1": "47"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $432 per year. It increases by 25% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "540", "answer_0": "530", "answer_2": "324", "answer_3": "546"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, December 15, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1621 per year. It increases by 25% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2025.25", "answer_2": "1215.75", "answer_0": "2009.25", "answer_1": "2026.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = 2026-05-15; priority = low.\n\nClaim to verify: \"the date is 2026-05-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $892. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.083", "p0": "0.167", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, June 16, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Monday, July 13, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed June 14, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 21", "date_0": "June 23", "date_3": "June 23", "date_1": "June 21"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $19k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.200", "p3": "0.050", "p2": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, February 05, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "25", "answer_2": "10", "answer_0": "35", "answer_1": "15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Wednesday, July 22, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.222", "p3": "0.750", "p1": "0.222", "p0": "0.778"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 10:00 to 14:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "5h00", "dur_0": "4h30", "dur_1": "4h45", "dur_3": "3h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Thursday, April 16, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed April 10, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 15", "date_0": "April 17", "date_3": "April 17", "date_2": "April 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 83 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "90", "answer_3": "96", "answer_2": "108", "answer_1": "80"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.571", "p3": "0.538", "p1": "0.429", "p2": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.500", "p0": "0.500", "p3": "0.429"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1732 per year. It increases by 25% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2145", "answer_3": "2133", "answer_1": "2165", "answer_2": "1299"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $31k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed May 16, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 25", "date_1": "May 24", "date_3": "May 26", "date_0": "May 27"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.200", "p1": "0.200", "p0": "0.800", "p3": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Tuesday, October 20, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $525. A fee of $169 is charged, then a deposit of $39 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "282", "answer_0": "311", "answer_3": "281", "answer_1": "336"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p3": "0.333", "p2": "0.571", "p0": "0.429"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed September 09, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 16", "date_0": "September 18", "date_2": "September 16", "date_3": "September 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, November 13, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed May 28, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 08", "date_0": "June 10", "date_1": "June 06", "date_2": "June 08"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = $25; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 15:00 to 19:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "3h15", "dur_1": "4h15", "dur_0": "3h30", "dur_2": "3h45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p0": "0.200", "p1": "0.250", "p3": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p0": "0.500", "p3": "0.125", "p1": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "77", "answer_3": "35", "answer_2": "42", "answer_1": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.467", "p0": "0.533", "p2": "0.467", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1961. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "64", "age_2": "66", "age_3": "63", "age_0": "65"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 09:00 to 09:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h15", "dur_1": "0h15", "dur_0": "-1h30", "dur_2": "-1h45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $144. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p0": "0.150", "p3": "0.050", "p1": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 485:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 572:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = 2026-08-24; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed February 04, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 12", "date_3": "February 12", "date_2": "February 14", "date_1": "February 10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.444", "p2": "0.556", "p3": "0.375", "p1": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 59 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "60", "answer_1": "68", "answer_2": "90", "answer_0": "72"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $222. A fee of $106 is charged, then a deposit of $20 arrives. A second fee of $53 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "88", "answer_1": "78", "answer_3": "90", "answer_0": "83"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, February 17, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $596. A fee of $178 is charged, then a deposit of $41 arrives. A second fee of $89 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "397", "answer_1": "388", "answer_0": "370", "answer_3": "379"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 282:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 136 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "49", "answer_1": "61", "answer_0": "64", "answer_3": "78"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 3 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = $335; priority = high.\n\nClaim to verify: \"the status is $335\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 270 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "140", "answer_1": "149", "answer_2": "128", "answer_3": "122"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 10 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1662 per year. It increases by 20% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1329.6", "answer_3": "2014.4", "answer_0": "1983.4", "answer_1": "1994.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1336 per year. It increases by 20% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1558.2", "answer_3": "1596.2", "answer_1": "1603.2", "answer_2": "1068.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 286 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_3": "112", "answer_1": "114", "answer_2": "131"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1968. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_2": "59", "age_3": "56", "age_1": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = 2026-09-15; priority = high.\n\nClaim to verify: \"the owner is 2026-09-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.417", "p0": "0.333", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Sunday, March 15, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 7 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "6", "answer_3": "21", "answer_0": "27", "answer_1": "13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $750. A fee of $104 is charged, then a deposit of $86 arrives. A second fee of $52 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "690", "answer_2": "684", "answer_0": "680", "answer_3": "672"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $571. A fee of $161 is charged, then a deposit of $77 arrives. A second fee of $80 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "407", "answer_1": "423", "answer_2": "388", "answer_3": "425"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $792 per year. It increases by 20% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "950.4", "answer_3": "897.4", "answer_0": "914.4", "answer_2": "633.6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 624:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = $72; priority = high.\n\nClaim to verify: \"the amount is $72\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p3": "0.600", "p2": "0.333", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 331:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $449 per year. It increases by 5% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "426.55", "answer_3": "399.45", "answer_1": "471.45", "answer_0": "424.45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 529:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed September 22, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 30", "date_2": "September 28", "date_1": "September 28", "date_3": "September 30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = closed; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.333", "p1": "0.417", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $332 per year. It increases by 25% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "249", "answer_1": "415", "answer_0": "399", "answer_3": "369"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $375. A fee of $170 is charged, then a deposit of $90 arrives. A second fee of $85 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "210", "answer_2": "214", "answer_1": "188", "answer_3": "225"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.500", "p3": "0.444", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = 2026-07-01; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p2": "0.500", "p3": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 418:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1976. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "47", "age_0": "48", "age_3": "50", "age_2": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = 2026-09-13; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "49", "answer_1": "21", "answer_0": "63", "answer_2": "14"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = $576; priority = high.\n\nClaim to verify: \"the status is $576\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $10k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_2": "24", "answer_1": "27", "answer_0": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $310 per year. It increases by 5% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "325.5", "answer_0": "310.5", "answer_3": "297.5", "answer_2": "294.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=851, unit_price=276, discount_pct=0, region=EU, sku_total=1127 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1127", "v2": "23487", "v3": "235727", "v0": "234876"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $366. A fee of $48 is charged, then a deposit of $30 arrives. A second fee of $24 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "324", "answer_1": "332", "answer_3": "315", "answer_2": "335"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 709:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1963. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "61", "age_2": "62", "age_3": "59", "age_1": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1977. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "49", "age_3": "47", "age_2": "50", "age_1": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_3": "21", "answer_1": "38", "answer_2": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1559 per year. It increases by 20% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1840.8", "answer_3": "1852.8", "answer_1": "1870.8", "answer_2": "1247.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1970. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "55", "age_2": "56", "age_3": "57", "age_1": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=811, unit_price=827, discount_pct=0, region=EU, sku_total=1638 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "669886", "v2": "67069", "v0": "670697", "v1": "1638"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.714", "p0": "0.286", "p1": "0.714", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.636", "p3": "0.300", "p1": "0.636", "p0": "0.364"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 10 people. A shopper bought a 70 gram box and a 700 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $274 per year. It increases by 20% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "317.8", "answer_0": "299.8", "answer_1": "328.8", "answer_2": "219.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 6 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 387 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "138", "answer_2": "122", "answer_0": "135", "answer_1": "138"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = 2026-04-09; priority = high.\n\nClaim to verify: \"the amount is 2026-04-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_1": "18", "answer_3": "8", "answer_2": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1990. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "33", "age_2": "35", "age_0": "34", "age_3": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, August 08, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=763, unit_price=746, discount_pct=0, region=EU, sku_total=1509 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "569198", "v3": "568435", "v1": "1509", "v2": "56919"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 121 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "98", "answer_1": "79", "answer_2": "71", "answer_0": "84"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 214 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "76", "answer_3": "77", "answer_2": "94", "answer_0": "90"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "62", "age_1": "63", "age_0": "64", "age_2": "65"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: status = 2026-12-28; priority = low.\n\nClaim to verify: \"the status is 2026-12-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed June 23, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 06", "date_2": "July 04", "date_3": "July 04", "date_1": "July 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.750", "p2": "0.250", "p3": "0.714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 537:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_2": "18", "answer_3": "30", "answer_1": "23"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 270 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "96", "answer_0": "100", "answer_3": "111", "answer_1": "95"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.111", "p1": "0.800", "p2": "0.800", "p0": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = 2026-11-10; priority = low.\n\nClaim to verify: \"the owner is 2026-11-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.417", "p0": "0.462", "p2": "0.538", "p1": "0.538"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An account starts at $291. A fee of $190 is charged, then a deposit of $49 arrives. A second fee of $95 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "64", "answer_0": "55", "answer_3": "37", "answer_2": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $534. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = $197; priority = high.\n\nClaim to verify: \"the date is $197\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = $115; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.667", "p3": "0.636", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.385", "p0": "0.429", "p1": "0.571", "p2": "0.571"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, November 14, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed June 21, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 30", "date_0": "June 30", "date_1": "June 28", "date_2": "June 28"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Wednesday, October 28, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 803:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p2": "0.050", "p3": "0.050", "p0": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Thursday, June 25, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $274. A fee of $79 is charged, then a deposit of $24 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "182", "answer_2": "197", "answer_1": "161", "answer_0": "180"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.462", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1982. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "43", "age_1": "41", "age_0": "42", "age_3": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 345 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "160", "answer_1": "174", "answer_3": "168", "answer_2": "171"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed June 07, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 20", "date_3": "June 18", "date_0": "June 18", "date_1": "June 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, March 20, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.375", "p3": "0.125", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = $565; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $13k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.333", "p1": "0.417", "p3": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $733. A fee of $53 is charged, then a deposit of $21 arrives. A second fee of $26 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "675", "answer_2": "682", "answer_3": "658", "answer_1": "651"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 11:15 to 13:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "2h30", "dur_1": "2h15", "dur_3": "1h15", "dur_0": "2h00"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1872 per year. It increases by 15% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2152.8", "answer_0": "2118.8", "answer_2": "1591.2", "answer_3": "2083.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.250", "p0": "0.750", "p3": "0.714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 624:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed July 23, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "August 05", "date_2": "August 03", "date_3": "August 03", "date_1": "August 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed August 15, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 20", "date_3": "August 22", "date_2": "August 19", "date_0": "August 21"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 10 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 6 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 177 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "85", "answer_0": "72", "answer_1": "76", "answer_2": "87"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $373. A fee of $76 is charged, then a deposit of $81 arrives. A second fee of $38 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "348", "answer_0": "340", "answer_2": "334", "answer_3": "370"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1321 per year. It increases by 25% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1603.25", "answer_3": "1574.25", "answer_1": "1651.25", "answer_2": "990.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed October 27, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 10", "date_2": "November 08", "date_3": "November 08", "date_1": "November 06"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "15", "answer_3": "9", "answer_0": "21", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=573, unit_price=904, discount_pct=0, region=EU, sku_total=1477 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1477", "v3": "518565", "v0": "517992", "v2": "51799"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A subscription costs $357 per year. It increases by 20% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "405.4", "answer_1": "428.4", "answer_2": "285.6", "answer_3": "437.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.625", "p0": "0.500", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1962. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "63", "age_0": "64", "age_2": "65", "age_3": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.444", "p2": "0.444", "p3": "0.500", "p0": "0.556"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "33", "answer_2": "18", "answer_1": "23", "answer_3": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.333", "p3": "0.625", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed February 06, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 12", "date_2": "February 10", "date_1": "February 10", "date_3": "February 12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=503, unit_price=874, discount_pct=0, region=EU, sku_total=1377 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1377", "v3": "439119", "v2": "43962", "v0": "439622"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = 2026-05-06; priority = high.\n\nClaim to verify: \"the owner is 2026-05-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1969. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "57", "age_1": "54", "age_2": "56", "age_0": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 227 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "13", "answer_0": "24", "answer_1": "42", "answer_3": "8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1958. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "66", "age_0": "67", "age_3": "65", "age_2": "68"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = 2026-07-21; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $440. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 2 grams of salt per serving and serves 8 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 359:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=540, unit_price=601, discount_pct=0, region=EU, sku_total=1141 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "324540", "v2": "32454", "v3": "325080", "v1": "1141"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $37k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed December 28, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 06", "date_2": "January 04", "date_0": "January 06", "date_1": "January 04"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "9", "answer_3": "18", "answer_0": "27", "answer_1": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = 2026-11-10; priority = high.\n\nClaim to verify: \"the date is 2026-11-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $885. A fee of $118 is charged, then a deposit of $66 arrives. A second fee of $59 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "774", "answer_3": "800", "answer_2": "801", "answer_1": "747"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 12:15 to 19:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "7h00", "dur_3": "6h00", "dur_0": "6h45", "dur_2": "7h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = 2026-09-13; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = $690; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.700", "p0": "0.300", "p3": "0.222", "p2": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = $527; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = 2026-10-12; priority = high.\n\nClaim to verify: \"the date is 2026-10-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "20", "answer_2": "10", "answer_3": "10", "answer_1": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $205. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = 2026-04-14; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = $245; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 07:00 to 10:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h30", "dur_2": "3h00", "dur_1": "3h15", "dur_3": "2h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1971. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "53", "age_1": "52", "age_3": "55", "age_2": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = $236; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, April 25, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1969. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "55", "age_1": "56", "age_0": "57", "age_2": "58"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Monday, May 25, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p1": "0.800", "p2": "0.800", "p3": "0.111"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 07:45 to 10:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "1h15", "dur_2": "1h45", "dur_1": "2h15", "dur_0": "1h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "45", "answer_1": "33", "answer_2": "30", "answer_3": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = 2026-04-24; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 06:00 to 12:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "6h30", "dur_3": "5h30", "dur_2": "6h45", "dur_0": "6h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1454 per year. It increases by 10% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1599.4", "answer_0": "1569.4", "answer_2": "1308.6", "answer_3": "1560.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $80. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 544:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $801. A fee of $114 is charged, then a deposit of $78 arrives. A second fee of $57 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "720", "answer_2": "686", "answer_1": "716", "answer_0": "708"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 271 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "175", "answer_2": "163", "answer_3": "160", "answer_1": "173"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $251 per year. It increases by 25% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "188.25", "answer_1": "313.75", "answer_3": "317.75", "answer_0": "300.75"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed December 18, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 30", "date_1": "December 28", "date_2": "January 03", "date_0": "January 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, April 05, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = pending; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1987. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "40", "age_0": "38", "age_2": "39", "age_1": "37"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 459:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $813. A fee of $165 is charged, then a deposit of $47 arrives. A second fee of $82 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "613", "answer_3": "602", "answer_2": "596", "answer_1": "599"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p1": "0.250", "p0": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $27k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 8 grams of salt per serving and serves 6 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=340, unit_price=914, discount_pct=0, region=EU, sku_total=1254 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1254", "v0": "310760", "v3": "310420", "v2": "31076"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed February 03, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 13", "date_2": "February 11", "date_3": "February 13", "date_1": "February 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 760:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed January 27, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 03", "date_3": "February 03", "date_1": "February 01", "date_2": "February 05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 8 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 55 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "259", "answer_1": "247", "answer_3": "236", "answer_0": "256"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=392, unit_price=420, discount_pct=0, region=EU, sku_total=812 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "812", "v2": "16464", "v0": "164640", "v3": "165032"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.667", "p3": "0.167", "p1": "0.833"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 413:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.800", "p0": "0.200", "p2": "0.800", "p3": "0.111"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1602 per year. It increases by 25% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1973.5", "answer_1": "2002.5", "answer_0": "1963.5", "answer_2": "1201.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $369. A fee of $31 is charged, then a deposit of $57 arrives. A second fee of $15 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "401", "answer_2": "394", "answer_3": "358", "answer_0": "380"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=148, unit_price=442, discount_pct=0, region=EU, sku_total=590 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "65416", "v2": "6541", "v3": "65564", "v1": "590"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=217, unit_price=663, discount_pct=0, region=EU, sku_total=880 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "14387", "v0": "143871", "v3": "144088", "v1": "880"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $16k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 756:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $768. A fee of $101 is charged, then a deposit of $21 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "647", "answer_3": "609", "answer_0": "638", "answer_1": "648"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.625", "p1": "0.625", "p0": "0.375", "p3": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = closed; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $715. A fee of $91 is charged, then a deposit of $43 arrives. A second fee of $45 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "604", "answer_1": "639", "answer_2": "610", "answer_0": "622"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 377 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "196", "answer_3": "179", "answer_2": "183", "answer_1": "186"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=723, unit_price=504, discount_pct=0, region=EU, sku_total=1227 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1227", "v3": "365115", "v2": "36439", "v0": "364392"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1968. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "57", "age_2": "58", "age_3": "59", "age_1": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = 2026-01-26; priority = high.\n\nClaim to verify: \"the date is 2026-01-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, November 21, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed April 13, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 26", "date_3": "April 24", "date_1": "April 22", "date_0": "April 24"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.083", "p0": "0.333", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.167", "p1": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, May 19, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 15:15 to 20:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "5h15", "dur_0": "5h00", "dur_3": "4h30", "dur_1": "5h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $750. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=649, unit_price=939, discount_pct=0, region=EU, sku_total=1588 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "609411", "v3": "610060", "v1": "1588", "v2": "60941"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.538", "p2": "0.462", "p1": "0.462", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 356:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = closed; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 254 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "168", "answer_3": "173", "answer_2": "148", "answer_1": "163"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = 2026-07-13; priority = low.\n\nClaim to verify: \"the amount is 2026-07-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed November 07, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 19", "date_1": "November 14", "date_3": "November 16", "date_0": "November 17"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 393 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "88", "answer_1": "79", "answer_2": "88", "answer_0": "72"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=637, unit_price=224, discount_pct=0, region=EU, sku_total=861 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "14268", "v1": "861", "v3": "143325", "v0": "142688"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed November 07, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 14", "date_1": "November 11", "date_3": "November 13", "date_0": "November 12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = ops; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.167", "p1": "0.333", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $353 per year. It increases by 15% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "335.95", "answer_1": "405.95", "answer_2": "300.05", "answer_0": "371.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = 2026-06-27; priority = low.\n\nClaim to verify: \"the amount is 2026-06-27\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 393 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "132", "answer_0": "120", "answer_3": "140", "answer_1": "102"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p3": "0.050", "p1": "0.250", "p0": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 9 grams of salt per serving and serves 8 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = $872; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 127 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "40", "answer_2": "39", "answer_3": "22", "answer_0": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.700", "p2": "0.700", "p3": "0.222", "p0": "0.300"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 312:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 773:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 359 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "331", "answer_0": "315", "answer_1": "303", "answer_3": "298"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1963. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "61", "age_3": "59", "age_2": "62", "age_1": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $376 per year. It increases by 25% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "410", "answer_2": "282", "answer_0": "435", "answer_1": "470"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_0": "60", "answer_2": "18", "answer_1": "25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $364 per year. It increases by 25% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "455", "answer_0": "413", "answer_3": "448", "answer_2": "273"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $658 per year. It increases by 5% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "690.9", "answer_0": "653.9", "answer_2": "625.1", "answer_3": "634.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.333", "p1": "0.500", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 09:45 to 18:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "8h15", "dur_0": "8h00", "dur_3": "7h15", "dur_2": "8h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $883. A fee of $58 is charged, then a deposit of $75 arrives. A second fee of $29 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "871", "answer_2": "841", "answer_1": "853", "answer_3": "898"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1961. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "66", "age_0": "65", "age_3": "67", "age_1": "64"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.429", "p1": "0.429", "p0": "0.571", "p3": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed January 14, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 29", "date_1": "January 23", "date_3": "January 25", "date_0": "January 27"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, September 08, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 859:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=962, unit_price=825, discount_pct=0, region=EU, sku_total=1787 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1787", "v0": "793650", "v3": "794612", "v2": "79365"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=532, unit_price=666, discount_pct=0, region=EU, sku_total=1198 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "353780", "v1": "1198", "v0": "354312", "v2": "35431"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.385", "p3": "0.333", "p2": "0.615", "p1": "0.615"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 2 grams of salt per serving and serves 10 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, May 19, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $250. A fee of $145 is charged, then a deposit of $47 arrives. A second fee of $72 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_1": "51", "answer_3": "102", "answer_2": "71"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p3": "0.125", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=519, unit_price=614, discount_pct=0, region=EU, sku_total=1133 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "318147", "v0": "318666", "v1": "1133", "v2": "31866"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An account starts at $516. A fee of $105 is charged, then a deposit of $81 arrives. A second fee of $52 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "440", "answer_2": "470", "answer_3": "419", "answer_1": "444"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: status = 2026-03-24; priority = low.\n\nClaim to verify: \"the status is 2026-03-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $797 per year. It increases by 5% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "796.85", "answer_2": "757.15", "answer_0": "821.85", "answer_1": "836.85"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "64", "answer_1": "22", "answer_2": "16", "answer_3": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.125", "p2": "0.250", "p0": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = 2026-09-07; priority = low.\n\nClaim to verify: \"the status is 2026-09-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $245. A fee of $148 is charged, then a deposit of $46 arrives. A second fee of $74 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "44", "answer_0": "69", "answer_1": "59", "answer_2": "89"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $29k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.200", "p1": "0.250", "p2": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1983. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "42", "age_2": "44", "age_0": "43", "age_3": "41"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.150", "p0": "0.200", "p3": "0.050"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $515. A fee of $179 is charged, then a deposit of $20 arrives. A second fee of $89 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "292", "answer_2": "294", "answer_1": "251", "answer_0": "267"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = $112; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = closed; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p3": "0.125", "p1": "0.375", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1672 per year. It increases by 20% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1993.4", "answer_1": "2006.4", "answer_3": "2024.4", "answer_2": "1337.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=414, unit_price=806, discount_pct=0, region=EU, sku_total=1220 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1220", "v0": "333684", "v3": "333270", "v2": "33368"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1063 per year. It increases by 25% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1307.75", "answer_1": "1328.75", "answer_2": "797.25", "answer_3": "1329.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed May 13, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 24", "date_1": "May 20", "date_3": "May 22", "date_0": "May 22"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed December 21, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 27", "date_0": "December 29", "date_3": "December 29", "date_1": "December 27"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.800", "p0": "0.200", "p2": "0.800", "p3": "0.111"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = 2026-07-22; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 262:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1733 per year. It increases by 5% this year, then a $40 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1779.65", "answer_3": "1816.65", "answer_2": "1646.35", "answer_1": "1819.65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1835 per year. It increases by 10% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1974.5", "answer_0": "1991.5", "answer_1": "2018.5", "answer_2": "1651.5"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_3": "18", "answer_1": "18", "answer_0": "30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed June 25, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "June 30", "date_2": "June 28", "date_1": "June 28", "date_3": "June 30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1979. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "47", "age_3": "44", "age_1": "45", "age_0": "46"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = $255; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 5 grams of salt per serving and serves 9 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.500", "p0": "0.375", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p2": "0.600", "p3": "0.333", "p0": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_1": "16", "answer_3": "12", "answer_0": "24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.167", "p1": "0.833", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = $847; priority = low.\n\nClaim to verify: \"the amount is $847\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1758 per year. It increases by 20% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2109.6", "answer_2": "1406.4", "answer_3": "2109.6", "answer_0": "2085.6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1972. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "55", "age_0": "54", "age_1": "53", "age_3": "52"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.455", "p0": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $692. A fee of $35 is charged, then a deposit of $79 arrives. A second fee of $17 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "719", "answer_3": "721", "answer_2": "705", "answer_1": "723"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.385", "p3": "0.583", "p0": "0.615", "p1": "0.385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $720. A fee of $152 is charged, then a deposit of $65 arrives. A second fee of $76 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "539", "answer_0": "557", "answer_3": "545", "answer_1": "535"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed March 18, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 30", "date_2": "March 30", "date_1": "March 28", "date_0": "April 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = ops; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 235 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "112", "answer_2": "131", "answer_3": "126", "answer_1": "116"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 16:45 to 17:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "0h30", "dur_1": "0h30", "dur_3": "-1h30", "dur_0": "0h15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = 2026-03-18; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, May 11, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed May 15, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 22", "date_3": "May 24", "date_2": "May 28", "date_0": "May 26"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "31", "answer_0": "80", "answer_2": "24", "answer_3": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.400", "p0": "0.600", "p3": "0.556", "p2": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, September 12, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $20k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1645 per year. It increases by 20% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1974", "answer_3": "1943", "answer_2": "1316", "answer_0": "1948"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 10:45 to 13:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "2h15", "dur_2": "2h30", "dur_3": "2h00", "dur_1": "3h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1597 per year. It increases by 25% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1197.75", "answer_3": "2005.25", "answer_1": "1996.25", "answer_0": "1968.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 117 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "230", "answer_2": "226", "answer_0": "210", "answer_1": "224"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 728:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 400:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.250", "p2": "0.125", "p1": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 780:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 4 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1912 per year. It increases by 15% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2198.8", "answer_0": "2188.8", "answer_3": "2174.8", "answer_2": "1625.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $616 per year. It increases by 10% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "626.6", "answer_2": "554.4", "answer_1": "677.6", "answer_0": "664.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = 2026-05-05; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 307:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.500", "p3": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p3": "0.050", "p0": "0.100", "p1": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=925, unit_price=396, discount_pct=0, region=EU, sku_total=1321 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "36630", "v1": "1321", "v0": "366300", "v3": "365375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.750", "p0": "0.778", "p1": "0.222", "p2": "0.222"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed January 17, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 01", "date_0": "January 30", "date_3": "January 29", "date_1": "January 27"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = $876; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Thursday, August 06, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed November 09, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 14", "date_0": "November 12", "date_2": "November 14", "date_1": "November 12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Monday, August 24, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $524. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 187 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_1": "65", "answer_2": "52", "answer_3": "91"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.667", "p3": "0.600", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Sunday, April 26, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed February 18, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 23", "date_2": "February 23", "date_3": "February 25", "date_0": "February 25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 8 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, March 10, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = 2026-03-16; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 12 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $477. A fee of $70 is charged, then a deposit of $54 arrives. A second fee of $35 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "426", "answer_2": "419", "answer_3": "411", "answer_1": "418"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = 2026-12-13; priority = low.\n\nClaim to verify: \"the status is 2026-12-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $200. A fee of $117 is charged, then a deposit of $42 arrives. A second fee of $58 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "67", "answer_3": "48", "answer_1": "53", "answer_2": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 935:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = $350; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed February 08, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 13", "date_2": "February 09", "date_1": "February 11", "date_0": "February 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1468 per year. It increases by 20% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1761.6", "answer_2": "1174.4", "answer_3": "1742.6", "answer_0": "1715.6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1212 per year. It increases by 20% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "969.6", "answer_0": "1411.4", "answer_1": "1454.4", "answer_3": "1400.4"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed June 13, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 16", "date_3": "June 18", "date_0": "June 17", "date_2": "June 15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 378 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "112", "answer_2": "128", "answer_1": "96", "answer_3": "118"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "25", "answer_3": "20", "answer_0": "45", "answer_1": "29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 364 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "60", "answer_1": "49", "answer_3": "47", "answer_0": "64"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=612, unit_price=193, discount_pct=0, region=EU, sku_total=805 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "118116", "v1": "805", "v3": "117504", "v2": "11811"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = ops; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1984. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "40", "age_3": "42", "age_1": "39", "age_2": "41"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = 2026-07-02; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = 2026-07-09; priority = low.\n\nClaim to verify: \"the owner is 2026-07-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = 2026-03-25; priority = low.\n\nClaim to verify: \"the status is 2026-03-25\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=552, unit_price=409, discount_pct=0, region=EU, sku_total=961 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "225768", "v2": "22576", "v1": "961", "v3": "225216"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.778", "p1": "0.222", "p3": "0.750", "p2": "0.222"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = $668; priority = high.\n\nClaim to verify: \"the date is $668\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "14", "answer_3": "8", "answer_0": "20", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $21k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $258. A fee of $92 is charged, then a deposit of $74 arrives. A second fee of $46 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "194", "answer_2": "189", "answer_3": "198", "answer_1": "200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.200", "p2": "0.100", "p0": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=255, unit_price=196, discount_pct=0, region=EU, sku_total=451 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "49725", "v0": "49980", "v1": "451", "v2": "4998"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 123 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "163", "answer_3": "150", "answer_0": "168", "answer_2": "153"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $264. A fee of $114 is charged, then a deposit of $29 arrives. A second fee of $57 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "137", "answer_0": "122", "answer_2": "97", "answer_3": "148"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = pending; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $594. A fee of $159 is charged, then a deposit of $77 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "433", "answer_1": "445", "answer_3": "426", "answer_2": "456"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = 2026-09-21; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed March 10, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 13", "date_1": "March 13", "date_2": "March 11", "date_3": "March 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.273", "p2": "0.273", "p3": "0.700", "p0": "0.727"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $376. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p0": "0.333", "p1": "0.667", "p3": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, January 17, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=219, unit_price=443, discount_pct=0, region=EU, sku_total=662 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "97017", "v2": "9701", "v3": "97236", "v1": "662"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 220 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "64", "answer_2": "103", "answer_1": "100", "answer_0": "84"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 442:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 553:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed July 13, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 19", "date_2": "July 15", "date_0": "July 17", "date_1": "July 17"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 3 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.375", "p3": "0.125", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed July 25, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 03", "date_0": "August 04", "date_2": "August 06", "date_1": "August 01"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 442:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p0": "0.333", "p3": "0.200", "p1": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $34k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = $534; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.429", "p3": "0.500", "p0": "0.571", "p2": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: status = closed; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.444", "p3": "0.500", "p0": "0.556", "p1": "0.444"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = 2026-10-06; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $783. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p3": "0.050", "p0": "0.200", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_3": "25", "answer_1": "20", "answer_2": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1974. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "49", "age_0": "51", "age_2": "52", "age_1": "50"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, October 20, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = $928; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 2 shifts this week. A client ordered 325 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "126", "answer_1": "114", "answer_3": "136", "answer_2": "138"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 14:30 to 20:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "5h15", "dur_2": "5h45", "dur_3": "4h30", "dur_1": "5h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 16:45 to 17:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "1h00", "dur_2": "0h45", "dur_0": "0h30", "dur_3": "0h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $620. A fee of $187 is charged, then a deposit of $32 arrives. A second fee of $93 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "378", "answer_1": "374", "answer_2": "400", "answer_0": "372"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = 2026-03-04; priority = low.\n\nClaim to verify: \"the amount is 2026-03-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.150", "p0": "0.100", "p2": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1983. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "44", "age_0": "43", "age_3": "45", "age_1": "42"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=397, unit_price=636, discount_pct=0, region=EU, sku_total=1033 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "252492", "v2": "25249", "v3": "252095", "v1": "1033"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1969. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "54", "age_0": "55", "age_3": "53", "age_2": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_1": "36", "answer_3": "36", "answer_0": "66"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = 2026-05-16; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $492. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.333", "p3": "0.167", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed December 27, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 04", "date_2": "January 06", "date_1": "January 02", "date_0": "January 04"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.167", "p1": "0.333", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 405:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p1": "0.625", "p0": "0.500", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 3 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $601. A fee of $78 is charged, then a deposit of $44 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "528", "answer_2": "523", "answer_3": "530", "answer_1": "544"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $775. A fee of $108 is charged, then a deposit of $45 arrives. A second fee of $54 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "645", "answer_2": "631", "answer_3": "638", "answer_0": "658"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $888 per year. It increases by 20% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1026.6", "answer_0": "1046.6", "answer_1": "1065.6", "answer_2": "710.4"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $93. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = $736; priority = high.\n\nClaim to verify: \"the amount is $736\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1973. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "50", "age_2": "52", "age_0": "51", "age_3": "53"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 401:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed October 25, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 05", "date_0": "November 03", "date_1": "November 01", "date_3": "November 03"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed October 27, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 07", "date_1": "November 05", "date_0": "November 09", "date_2": "November 11"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.083", "p3": "0.083", "p0": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = 2026-05-24; priority = high.\n\nClaim to verify: \"the date is 2026-05-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 07:15 to 08:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h00", "dur_3": "0h15", "dur_1": "1h15", "dur_0": "0h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed February 22, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 03", "date_2": "March 03", "date_0": "March 05", "date_3": "March 05"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "15", "answer_1": "9", "answer_2": "6", "answer_3": "9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p3": "0.467", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 229 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "144", "answer_3": "137", "answer_1": "142", "answer_2": "148"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = 2026-02-28; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1983. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "41", "age_2": "42", "age_1": "40", "age_3": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed September 23, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 29", "date_0": "September 29", "date_2": "October 01", "date_1": "September 27"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=721, unit_price=453, discount_pct=0, region=EU, sku_total=1174 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1174", "v2": "32661", "v3": "325892", "v0": "326613"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1990. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "36", "age_3": "33", "age_1": "34", "age_0": "35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "36", "answer_0": "72", "answer_1": "42", "answer_3": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "14", "answer_0": "35", "answer_1": "17", "answer_3": "21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "32", "answer_3": "56", "answer_1": "39", "answer_0": "88"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = $480; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p2": "0.100", "p3": "0.050", "p0": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1041 per year. It increases by 5% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1073.05", "answer_0": "1083.05", "answer_2": "988.95", "answer_1": "1093.05"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = $68; priority = high.\n\nClaim to verify: \"the date is $68\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 6 grams of salt per serving and serves 5 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.667", "p0": "0.500", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $747 per year. It increases by 10% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "672.3", "answer_0": "778.7", "answer_1": "821.7", "answer_3": "804.7"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.600", "p2": "0.364", "p0": "0.636", "p1": "0.364"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Wednesday, March 04, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "16", "answer_2": "12", "answer_0": "24", "answer_3": "12"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.538", "p0": "0.462", "p3": "0.417", "p1": "0.538"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = 2026-11-19; priority = low.\n\nClaim to verify: \"the amount is 2026-11-19\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1990. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "33", "age_0": "34", "age_3": "36", "age_2": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $238. A fee of $31 is charged, then a deposit of $52 arrives. A second fee of $15 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "244", "answer_3": "230", "answer_1": "228", "answer_2": "239"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1971. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "53", "age_0": "54", "age_2": "55", "age_3": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = $270; priority = high.\n\nClaim to verify: \"the owner is $270\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Monday, February 16, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 436:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1979. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "46", "age_3": "44", "age_1": "45", "age_2": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, May 03, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 7 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 155 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "126", "answer_3": "154", "answer_1": "134", "answer_0": "140"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed February 21, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 26", "date_2": "February 25", "date_0": "February 27", "date_3": "February 28"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1977. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "49", "age_3": "46", "age_1": "47", "age_0": "48"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.545", "p0": "0.455", "p3": "0.400", "p1": "0.545"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = $511; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: owner = 2026-06-18; priority = high.\n\nClaim to verify: \"the owner is 2026-06-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 506:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 5 grams of salt per serving and serves 7 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = 2026-01-13; priority = high.\n\nClaim to verify: \"the date is 2026-01-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $7k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1968. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "57", "age_2": "58", "age_3": "59", "age_1": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=522, unit_price=891, discount_pct=0, region=EU, sku_total=1413 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "465102", "v2": "46510", "v1": "1413", "v3": "465624"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 283 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "205", "answer_3": "235", "answer_0": "216", "answer_2": "211"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "62", "age_0": "64", "age_2": "65", "age_1": "63"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.050", "p0": "0.200", "p2": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $431 per year. It increases by 20% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "519.2", "answer_2": "344.8", "answer_0": "492.2", "answer_1": "517.2"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 988:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 06:45 to 08:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h00", "dur_3": "1h00", "dur_0": "1h45", "dur_2": "2h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed March 24, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 27", "date_0": "March 27", "date_3": "March 29", "date_2": "March 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = 2026-09-13; priority = low.\n\nClaim to verify: \"the owner is 2026-09-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_3": "15", "answer_1": "14", "answer_2": "9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_0": "60", "answer_3": "30", "answer_1": "35"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $524 per year. It increases by 20% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "419.2", "answer_1": "628.8", "answer_3": "616.8", "answer_0": "595.8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = 2026-07-14; priority = high.\n\nClaim to verify: \"the date is 2026-07-14\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=850, unit_price=694, discount_pct=0, region=EU, sku_total=1544 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "589900", "v1": "1544", "v3": "590750", "v2": "58990"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 419:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 970:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p2": "0.500", "p3": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 214 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "230", "answer_3": "230", "answer_1": "194", "answer_0": "210"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 383 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "21", "answer_0": "36", "answer_2": "17", "answer_1": "22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $596. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.429", "p1": "0.429", "p0": "0.571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "32", "answer_0": "42", "answer_3": "12", "answer_2": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=170, unit_price=439, discount_pct=0, region=EU, sku_total=609 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "74460", "v0": "74630", "v1": "609", "v2": "7463"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.500", "p2": "0.167", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.667", "p3": "0.167", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.429", "p1": "0.429", "p0": "0.571", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Monday, July 20, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, September 20, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = $903; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $417. A fee of $199 is charged, then a deposit of $66 arrives. A second fee of $99 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "171", "answer_2": "191", "answer_1": "175", "answer_0": "185"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 489:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $578. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Sunday, January 04, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "54", "answer_2": "12", "answer_1": "19", "answer_3": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 63 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "215", "answer_1": "216", "answer_0": "210", "answer_2": "212"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 124 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "20", "answer_0": "36", "answer_1": "48", "answer_3": "22"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = 2026-07-15; priority = low.\n\nClaim to verify: \"the date is 2026-07-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $773. A fee of $179 is charged, then a deposit of $58 arrives. A second fee of $89 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "571", "answer_2": "587", "answer_3": "560", "answer_0": "563"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, September 26, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed April 01, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 05", "date_0": "April 07", "date_3": "April 07", "date_1": "April 05"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=272, unit_price=681, discount_pct=0, region=EU, sku_total=953 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "953", "v2": "18523", "v3": "185504", "v0": "185232"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = 2026-07-26; priority = high.\n\nClaim to verify: \"the date is 2026-07-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "41", "answer_3": "30", "answer_0": "66", "answer_2": "36"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $721; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p2": "0.200", "p3": "0.778", "p0": "0.800"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = $747; priority = high.\n\nClaim to verify: \"the amount is $747\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $351. A fee of $149 is charged, then a deposit of $87 arrives. A second fee of $74 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "210", "answer_2": "209", "answer_0": "215", "answer_3": "226"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 18:45 to 20:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h45", "dur_3": "1h00", "dur_1": "2h00", "dur_0": "1h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 855:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "42", "answer_2": "36", "answer_0": "72", "answer_3": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p1": "0.625", "p0": "0.500", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 483:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 5 grams of salt per serving and serves 4 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 10:30 to 16:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "4h45", "dur_0": "5h15", "dur_2": "5h30", "dur_1": "5h45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=467, unit_price=899, discount_pct=0, region=EU, sku_total=1366 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "419833", "v2": "41983", "v1": "1366", "v3": "420300"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1138 per year. It increases by 25% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1358.5", "answer_1": "1422.5", "answer_0": "1391.5", "answer_2": "853.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $216. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p3": "0.050", "p0": "0.100", "p1": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 16:45 to 21:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "3h45", "dur_2": "4h15", "dur_1": "4h45", "dur_0": "4h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.667", "p3": "0.600", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_1": "22", "answer_3": "48", "answer_0": "64"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.375", "p2": "0.250", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = 2026-10-09; priority = low.\n\nClaim to verify: \"the owner is 2026-10-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $578. A fee of $62 is charged, then a deposit of $46 arrives. A second fee of $31 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "511", "answer_2": "527", "answer_1": "536", "answer_0": "531"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 175 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "165", "answer_1": "183", "answer_3": "195", "answer_0": "180"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = 2026-08-22; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 6 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.625", "p0": "0.500", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed March 01, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 15", "date_0": "March 13", "date_1": "March 11", "date_3": "March 13"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed June 27, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 05", "date_0": "July 07", "date_1": "July 04", "date_3": "July 06"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $220. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 162 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "85", "answer_1": "83", "answer_0": "96", "answer_3": "99"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 415:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $265. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.444", "p1": "0.500", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.050", "p1": "0.150", "p0": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $885. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 10:45 to 19:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "8h30", "dur_3": "8h00", "dur_2": "9h00", "dur_1": "9h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=567, unit_price=191, discount_pct=0, region=EU, sku_total=758 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "108864", "v0": "108297", "v1": "758", "v2": "10829"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $756. A fee of $89 is charged, then a deposit of $68 arrives. A second fee of $44 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "691", "answer_1": "721", "answer_2": "693", "answer_3": "698"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $34k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=797, unit_price=797, discount_pct=0, region=EU, sku_total=1594 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "63520", "v0": "635209", "v3": "636006", "v1": "1594"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.727", "p2": "0.727", "p3": "0.200", "p0": "0.273"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.083", "p2": "0.083", "p0": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1239 per year. It increases by 10% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1115.1", "answer_1": "1362.9", "answer_3": "1306.9", "answer_0": "1346.9"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed February 18, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 26", "date_2": "February 24", "date_1": "February 24", "date_0": "February 26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = 2026-10-17; priority = high.\n\nClaim to verify: \"the amount is 2026-10-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1618 per year. It increases by 5% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1698.9", "answer_0": "1680.9", "answer_3": "1652.9", "answer_2": "1537.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = $593; priority = high.\n\nClaim to verify: \"the amount is $593\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $293. A fee of $76 is charged, then a deposit of $25 arrives. A second fee of $38 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "224", "answer_0": "204", "answer_2": "226", "answer_1": "230"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.333", "p0": "0.500", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $38k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.667", "p1": "0.333", "p3": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "8", "answer_0": "24", "answer_1": "12", "answer_3": "16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = $371; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 14:30 to 18:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h45", "dur_3": "2h45", "dur_0": "3h15", "dur_1": "3h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Wednesday, December 02, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1968. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "57", "age_1": "55", "age_0": "56", "age_3": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 581:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = $752; priority = high.\n\nClaim to verify: \"the status is $752\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 239 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "269", "answer_3": "276", "answer_2": "286", "answer_0": "280"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed April 13, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 20", "date_2": "April 18", "date_1": "April 18", "date_0": "April 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed August 28, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 02", "date_2": "August 31", "date_3": "September 02", "date_1": "August 31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed October 10, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 19", "date_0": "October 22", "date_2": "October 24", "date_3": "October 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=842, unit_price=342, discount_pct=0, region=EU, sku_total=1184 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "288806", "v2": "28796", "v0": "287964", "v1": "1184"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, August 10, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.250", "p0": "0.167", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 699:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 224 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "63", "answer_3": "58", "answer_0": "72", "answer_2": "91"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p1": "0.625", "p3": "0.286", "p2": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: owner = $836; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "53", "age_3": "50", "age_0": "52", "age_1": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 121 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "205", "answer_3": "196", "answer_2": "212", "answer_0": "210"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 77 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "22", "answer_3": "7", "answer_0": "24", "answer_2": "41"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $857. A fee of $133 is charged, then a deposit of $86 arrives. A second fee of $66 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "723", "answer_0": "744", "answer_2": "762", "answer_3": "765"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 205 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "126", "answer_2": "146", "answer_1": "131", "answer_3": "140"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 6 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1988. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "37", "age_3": "35", "age_1": "36", "age_2": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $3k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $899. A fee of $54 is charged, then a deposit of $39 arrives. A second fee of $27 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "834", "answer_3": "886", "answer_0": "857", "answer_1": "830"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed January 28, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "January 31", "date_0": "February 02", "date_3": "February 02", "date_1": "January 31"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=643, unit_price=415, discount_pct=0, region=EU, sku_total=1058 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "266845", "v2": "26684", "v3": "266202", "v1": "1058"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed April 15, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 23", "date_1": "April 21", "date_3": "April 23", "date_2": "April 25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 5 people. A shopper bought a 25 gram box and a 250 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 7 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, September 25, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, December 08, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 441:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $33k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 287 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "132", "answer_2": "120", "answer_3": "103", "answer_0": "112"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 326 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "212", "answer_0": "196", "answer_3": "177", "answer_1": "183"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p3": "0.167", "p0": "0.667", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1975. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "53", "age_1": "50", "age_2": "52", "age_0": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, March 08, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1984. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "42", "age_2": "41", "age_0": "40", "age_1": "39"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.250", "p0": "0.750", "p3": "0.714"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = 2026-02-15; priority = high.\n\nClaim to verify: \"the status is 2026-02-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p3": "0.429", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $570. A fee of $107 is charged, then a deposit of $78 arrives. A second fee of $53 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "479", "answer_1": "503", "answer_0": "488", "answer_3": "470"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p2": "0.600", "p3": "0.250", "p1": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, January 03, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 3 grams of salt per serving and serves 8 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = 2026-10-23; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1982. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "44", "age_0": "42", "age_2": "43", "age_1": "41"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 6 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p2": "0.600", "p3": "0.333", "p0": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = 2026-03-15; priority = high.\n\nClaim to verify: \"the amount is 2026-03-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 933:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "20", "answer_1": "27", "answer_0": "55", "answer_3": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=801, unit_price=127, discount_pct=0, region=EU, sku_total=928 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "10172", "v0": "101727", "v1": "928", "v3": "100926"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 136 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "56", "answer_0": "60", "answer_2": "80", "answer_3": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $380. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed March 23, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 02", "date_0": "March 31", "date_1": "March 29", "date_3": "March 31"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.571", "p0": "0.429", "p1": "0.571", "p3": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 137 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "38", "answer_0": "40", "answer_3": "46", "answer_1": "60"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=927, unit_price=253, discount_pct=0, region=EU, sku_total=1180 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "23453", "v0": "234531", "v1": "1180", "v3": "235458"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1983. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "41", "age_1": "42", "age_2": "44", "age_0": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.250", "p1": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=821, unit_price=946, discount_pct=0, region=EU, sku_total=1767 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1767", "v2": "77666", "v0": "776666", "v3": "775845"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 236 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "120", "answer_3": "147", "answer_0": "140", "answer_2": "144"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.222", "p0": "0.300", "p1": "0.700", "p2": "0.700"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 4 grams of salt per serving and serves 5 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $39k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.615", "p0": "0.385", "p3": "0.333", "p2": "0.615"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = 2026-12-04; priority = high.\n\nClaim to verify: \"the date is 2026-12-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = $643; priority = high.\n\nClaim to verify: \"the amount is $643\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $3k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 5 grams of salt per serving and serves 8 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 799:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 12:45 to 22:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "8h30", "dur_2": "9h15", "dur_1": "9h30", "dur_0": "9h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.400", "p1": "0.400", "p0": "0.600", "p3": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Thursday, June 04, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = $921; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 81 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "147", "answer_1": "167", "answer_3": "132", "answer_2": "153"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 5 grams of salt per serving and serves 7 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.250", "p1": "0.375", "p2": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p0": "0.333", "p2": "0.250", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = ops; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Thursday, April 09, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 123 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "74", "answer_1": "84", "answer_3": "85", "answer_0": "72"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.125", "p2": "0.375", "p1": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 918:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "20", "answer_1": "34", "answer_0": "50", "answer_2": "30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1990. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "33", "age_0": "34", "age_2": "35", "age_3": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $772. A fee of $42 is charged, then a deposit of $40 arrives. A second fee of $21 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "763", "answer_1": "720", "answer_3": "742", "answer_0": "749"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.556", "p0": "0.600", "p2": "0.400", "p1": "0.400"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, April 28, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.714", "p0": "0.750", "p2": "0.250", "p1": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.583", "p1": "0.385", "p0": "0.615", "p2": "0.385"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.500", "p2": "0.333", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = 2026-01-21; priority = high.\n\nClaim to verify: \"the amount is 2026-01-21\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed July 20, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 23", "date_3": "July 25", "date_0": "July 23", "date_2": "July 21"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 326 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "65", "answer_0": "70", "answer_3": "61", "answer_1": "89"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=412, unit_price=476, discount_pct=0, region=EU, sku_total=888 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "196112", "v1": "888", "v2": "19611", "v3": "196524"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $794. A fee of $155 is charged, then a deposit of $29 arrives. A second fee of $77 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "569", "answer_1": "580", "answer_0": "591", "answer_3": "611"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_1": "21", "answer_3": "40", "answer_2": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $582. A fee of $65 is charged, then a deposit of $69 arrives. A second fee of $32 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "563", "answer_0": "554", "answer_2": "531", "answer_3": "583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = 2026-03-07; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=339, unit_price=326, discount_pct=0, region=EU, sku_total=665 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "110853", "v2": "11051", "v0": "110514", "v1": "665"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.600", "p1": "0.400", "p3": "0.556", "p2": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "56", "answer_0": "96", "answer_1": "47", "answer_2": "40"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p3": "0.625", "p1": "0.333", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = $137; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 874:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $5k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 21, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 26", "date_0": "April 28", "date_3": "April 28", "date_2": "April 30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p3": "0.050", "p1": "0.200", "p2": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $24k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = 2026-12-12; priority = low.\n\nClaim to verify: \"the date is 2026-12-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.200", "p2": "0.727", "p1": "0.727", "p0": "0.273"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 07:15 to 18:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "11h15", "dur_0": "11h00", "dur_1": "11h30", "dur_3": "10h30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $499. A fee of $199 is charged, then a deposit of $54 arrives. A second fee of $99 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "255", "answer_2": "237", "answer_1": "276", "answer_3": "238"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1322 per year. It increases by 20% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1057.6", "answer_3": "1605.4", "answer_1": "1586.4", "answer_0": "1567.4"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = 2026-09-27; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = $509; priority = high.\n\nClaim to verify: \"the date is $509\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p1": "0.250", "p2": "0.083", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, December 23, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $396. A fee of $66 is charged, then a deposit of $20 arrives. A second fee of $33 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "303", "answer_3": "315", "answer_0": "317", "answer_2": "292"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 894:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "15", "answer_1": "21", "answer_0": "33", "answer_3": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p0": "0.500", "p3": "0.125", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $3k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1976. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "49", "age_2": "50", "age_1": "48", "age_3": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = 2026-09-22; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1883 per year. It increases by 20% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1506.4", "answer_0": "2242.6", "answer_1": "2259.6", "answer_3": "2273.6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=838, unit_price=880, discount_pct=0, region=EU, sku_total=1718 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "73744", "v1": "1718", "v0": "737440", "v3": "736602"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = open; priority = high.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 8 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.167", "p3": "0.083", "p2": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $456 per year. It increases by 5% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "478.8", "answer_3": "468.8", "answer_0": "444.8", "answer_2": "433.2"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 7 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed January 11, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 20", "date_2": "January 18", "date_0": "January 20", "date_1": "January 18"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $850. A fee of $125 is charged, then a deposit of $88 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "761", "answer_1": "732", "answer_3": "721", "answer_0": "751"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $39k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Tuesday, August 04, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1264 per year. It increases by 20% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1516.8", "answer_0": "1505.8", "answer_3": "1516.8", "answer_2": "1011.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=662, unit_price=897, discount_pct=0, region=EU, sku_total=1559 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "593152", "v1": "1559", "v2": "59381", "v0": "593814"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 203:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 158 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "33", "answer_0": "48", "answer_1": "46", "answer_3": "28"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 908:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Tuesday, March 24, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = 2026-03-26; priority = low.\n\nClaim to verify: \"the amount is 2026-03-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: date = $620; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Thursday, August 20, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1982. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "42", "age_1": "43", "age_0": "44", "age_2": "45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1975. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "50", "age_2": "51", "age_1": "49", "age_3": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 3 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $453. A fee of $193 is charged, then a deposit of $26 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "190", "answer_2": "203", "answer_1": "173", "answer_3": "187"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.111", "p0": "0.200", "p1": "0.800", "p2": "0.800"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 561:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = 2026-10-03; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $50. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.083", "p0": "0.333", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = $953; priority = low.\n\nClaim to verify: \"the amount is $953\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 942:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $613. A fee of $151 is charged, then a deposit of $64 arrives. A second fee of $75 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "447", "answer_3": "478", "answer_1": "431", "answer_0": "451"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = $16; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $25k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.300", "p2": "0.636", "p1": "0.636", "p0": "0.364"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 19:45 to 22:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "2h30", "dur_1": "2h30", "dur_0": "2h15", "dur_3": "1h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $339. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $920 per year. It increases by 25% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "690", "answer_0": "1128", "answer_3": "1151", "answer_1": "1150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 145 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "133", "answer_2": "116", "answer_3": "113", "answer_0": "126"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p1": "0.833", "p3": "0.167", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1923 per year. It increases by 20% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2307.6", "answer_2": "1538.4", "answer_0": "2264.6", "answer_3": "2237.6"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 9 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, December 01, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 12 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 8 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1986. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "40", "age_3": "41", "age_0": "39", "age_1": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 14:15 to 19:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "5h15", "dur_3": "4h15", "dur_0": "5h00", "dur_2": "5h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 122 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "168", "answer_1": "155", "answer_2": "187", "answer_3": "184"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = 2026-10-22; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $35k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Wednesday, June 10, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=625, unit_price=245, discount_pct=0, region=EU, sku_total=870 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "15312", "v1": "870", "v3": "152500", "v0": "153125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 6 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $433. A fee of $148 is charged, then a deposit of $89 arrives. A second fee of $74 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "300", "answer_2": "282", "answer_3": "284", "answer_1": "298"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 137 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "75", "answer_1": "57", "answer_2": "66", "answer_3": "82"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 106 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "231", "answer_0": "224", "answer_2": "239", "answer_1": "220"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 08:30 to 11:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "3h00", "dur_0": "2h30", "dur_3": "2h15", "dur_1": "3h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $584 per year. It increases by 25% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "730", "answer_2": "438", "answer_3": "724", "answer_0": "717"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $477. A fee of $41 is charged, then a deposit of $36 arrives. A second fee of $20 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "434", "answer_3": "454", "answer_1": "426", "answer_0": "452"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed April 04, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 10", "date_0": "April 08", "date_1": "April 07", "date_3": "April 09"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.667", "p1": "0.833", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 374 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "76", "answer_0": "72", "answer_2": "89", "answer_1": "57"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 7 grams of salt per serving and serves 7 people. A shopper bought a 49 gram box and a 490 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = 2026-09-22; priority = low.\n\nClaim to verify: \"the amount is 2026-09-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 4 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 279:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1521 per year. It increases by 5% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1569.05", "answer_2": "1444.95", "answer_3": "1540.05", "answer_1": "1597.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.083", "p0": "0.167", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.333", "p2": "0.167", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=634, unit_price=564, discount_pct=0, region=EU, sku_total=1198 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1198", "v2": "35757", "v3": "356942", "v0": "357576"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.400", "p2": "0.500", "p0": "0.500", "p1": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $734 per year. It increases by 25% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "898.5", "answer_1": "917.5", "answer_2": "550.5", "answer_3": "866.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.429", "p2": "0.500", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = $931; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $357. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $13k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p2": "0.600", "p0": "0.400", "p3": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $894. A fee of $153 is charged, then a deposit of $38 arrives. A second fee of $76 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "718", "answer_0": "703", "answer_1": "691", "answer_3": "713"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $781. A fee of $146 is charged, then a deposit of $49 arrives. A second fee of $73 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "587", "answer_2": "626", "answer_0": "611", "answer_3": "607"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 283:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $328. A fee of $197 is charged, then a deposit of $39 arrives. A second fee of $98 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "84", "answer_1": "54", "answer_3": "101", "answer_0": "72"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1982. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "43", "age_1": "41", "age_0": "42", "age_3": "44"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "32", "answer_1": "35", "answer_0": "56", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.167", "p3": "0.083", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = 2026-02-21; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=566, unit_price=487, discount_pct=0, region=EU, sku_total=1053 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1053", "v0": "275642", "v3": "275076", "v2": "27564"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed March 24, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 02", "date_0": "April 02", "date_1": "March 31", "date_2": "April 04"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $345; priority = high.\n\nClaim to verify: \"the owner is $345\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "25", "answer_0": "35", "answer_1": "27", "answer_3": "10"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "24", "answer_3": "21", "answer_0": "42", "answer_2": "21"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $440. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.286", "p0": "0.375", "p2": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1635 per year. It increases by 5% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1553.25", "answer_0": "1695.75", "answer_3": "1714.75", "answer_1": "1716.75"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=161, unit_price=217, discount_pct=0, region=EU, sku_total=378 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "35098", "v0": "34937", "v1": "378", "v2": "3493"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1975. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "50", "age_2": "51", "age_3": "48", "age_1": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 967:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 162 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "240", "answer_2": "259", "answer_1": "242", "answer_0": "252"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 149 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "164", "answer_3": "145", "answer_0": "160", "answer_2": "180"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "11", "answer_2": "8", "answer_0": "20", "answer_3": "12"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1873 per year. It increases by 5% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1920.65", "answer_3": "1957.65", "answer_1": "1966.65", "answer_2": "1779.35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1468 per year. It increases by 10% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1583.8", "answer_3": "1549.8", "answer_2": "1321.2", "answer_1": "1614.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=784, unit_price=132, discount_pct=0, region=EU, sku_total=916 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "102704", "v0": "103488", "v1": "916", "v2": "10348"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = 2026-12-10; priority = low.\n\nClaim to verify: \"the status is 2026-12-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.100", "p1": "0.150", "p2": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.714", "p2": "0.250", "p1": "0.250", "p0": "0.750"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed December 23, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 06", "date_2": "January 08", "date_1": "January 02", "date_3": "January 04"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.462", "p1": "0.500", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1966. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "57", "age_2": "59", "age_0": "58", "age_3": "56"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed April 10, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 19", "date_2": "April 23", "date_1": "April 17", "date_0": "April 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = $262; priority = low.\n\nClaim to verify: \"the owner is $262\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.083", "p2": "0.250", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = 2026-10-22; priority = high.\n\nClaim to verify: \"the status is 2026-10-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, May 20, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = closed; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=619, unit_price=181, discount_pct=0, region=EU, sku_total=800 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "800", "v2": "11203", "v3": "112658", "v0": "112039"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p3": "0.050", "p2": "0.150", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $539; priority = high.\n\nClaim to verify: \"the owner is $539\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 320 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "207", "answer_1": "206", "answer_0": "216", "answer_3": "222"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_1": "15", "answer_3": "28", "answer_2": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed May 26, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 01", "date_0": "June 03", "date_2": "June 05", "date_3": "June 03"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p2": "0.100", "p1": "0.200", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = 2026-09-16; priority = low.\n\nClaim to verify: \"the status is 2026-09-16\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.500", "p3": "0.462", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = $871; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.300", "p0": "0.364", "p1": "0.636", "p2": "0.636"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=700, unit_price=583, discount_pct=0, region=EU, sku_total=1283 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1283", "v3": "408800", "v0": "408100", "v2": "40810"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 5 grams of salt per serving and serves 4 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=722, unit_price=383, discount_pct=0, region=EU, sku_total=1105 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "276526", "v2": "27652", "v1": "1105", "v3": "275804"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p3": "0.167", "p1": "0.667", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.200", "p2": "0.667", "p0": "0.333", "p1": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed June 17, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "June 23", "date_2": "June 21", "date_3": "June 23", "date_1": "June 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = open; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=738, unit_price=568, discount_pct=0, region=EU, sku_total=1306 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "418446", "v0": "419184", "v2": "41918", "v1": "1306"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $38k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.273", "p1": "0.667", "p0": "0.333", "p2": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1983. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "41", "age_1": "40", "age_2": "42", "age_3": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $642. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = $134; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 3 grams of salt per serving and serves 9 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = 2026-02-21; priority = high.\n\nClaim to verify: \"the owner is 2026-02-21\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1969. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "55", "age_0": "56", "age_3": "58", "age_2": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.462", "p1": "0.500", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 918:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_1": "29", "answer_0": "54", "answer_3": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p3": "0.050", "p2": "0.050", "p1": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = 2026-01-13; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A shift runs from 17:15 to 19:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "1h15", "dur_2": "2h15", "dur_1": "2h15", "dur_0": "1h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 364:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, January 25, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 101 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "100", "answer_3": "109", "answer_0": "120", "answer_2": "114"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1154 per year. It increases by 10% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1038.6", "answer_1": "1269.4", "answer_3": "1247.4", "answer_0": "1253.4"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.500", "p1": "0.833", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=542, unit_price=829, discount_pct=0, region=EU, sku_total=1371 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "449318", "v2": "44931", "v1": "1371", "v3": "448776"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p3": "0.083", "p2": "0.250", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=579, unit_price=892, discount_pct=0, region=EU, sku_total=1471 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "516468", "v3": "517047", "v2": "51646", "v1": "1471"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 287:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=261, unit_price=682, discount_pct=0, region=EU, sku_total=943 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "178263", "v2": "17800", "v0": "178002", "v1": "943"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 153 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "263", "answer_1": "284", "answer_0": "280", "answer_3": "263"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1962. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "62", "age_2": "65", "age_1": "63", "age_0": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.400", "p0": "0.455", "p1": "0.545", "p2": "0.545"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $339. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1968. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "59", "age_0": "57", "age_1": "56", "age_2": "58"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1518 per year. It increases by 10% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1669.8", "answer_0": "1643.8", "answer_3": "1610.8", "answer_2": "1366.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 66 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_2": "79", "answer_1": "75", "answer_3": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $23k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $13k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = 2026-03-15; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: date = closed; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 223:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.200", "p1": "0.200", "p0": "0.800", "p3": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=251, unit_price=628, discount_pct=0, region=EU, sku_total=879 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "157879", "v1": "879", "v0": "157628", "v2": "15762"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, August 03, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1176 per year. It increases by 5% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1204.8", "answer_2": "1117.2", "answer_1": "1234.8", "answer_3": "1238.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 4 grams of salt per serving and serves 12 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p3": "0.050", "p2": "0.100", "p0": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1987. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "39", "age_0": "38", "age_3": "40", "age_1": "37"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.333", "p3": "0.167", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 11:45 to 21:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "10h15", "dur_3": "9h00", "dur_1": "10h00", "dur_0": "9h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 4 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=278, unit_price=470, discount_pct=0, region=EU, sku_total=748 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "748", "v3": "130938", "v2": "13066", "v0": "130660"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 7 people. A shopper bought a 14 gram box and a 140 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=117, unit_price=927, discount_pct=0, region=EU, sku_total=1044 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "10845", "v0": "108459", "v1": "1044", "v3": "108342"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.778", "p1": "0.222", "p2": "0.222", "p3": "0.750"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1976. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "49", "age_1": "47", "age_0": "48", "age_3": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=175, unit_price=153, discount_pct=0, region=EU, sku_total=328 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "2677", "v1": "328", "v0": "26775", "v3": "26600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 3 grams of salt per serving and serves 3 people. A shopper bought a 9 gram box and a 90 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 614:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, September 11, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = $505; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $563. A fee of $59 is charged, then a deposit of $34 arrives. A second fee of $29 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "489", "answer_2": "531", "answer_3": "525", "answer_0": "509"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $526. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.333", "p3": "0.083", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = ops; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed November 16, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 26", "date_0": "November 30", "date_3": "November 28", "date_2": "November 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = 2026-03-26; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1961. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "63", "age_3": "61", "age_2": "64", "age_1": "62"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $732. A fee of $121 is charged, then a deposit of $72 arrives. A second fee of $60 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "595", "answer_3": "606", "answer_2": "648", "answer_0": "623"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 528:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1968. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "56", "age_1": "55", "age_3": "54", "age_2": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 5 grams of salt per serving and serves 5 people. A shopper bought a 25 gram box and a 250 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $870. A fee of $68 is charged, then a deposit of $38 arrives. A second fee of $34 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "821", "answer_3": "835", "answer_2": "832", "answer_0": "806"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=389, unit_price=220, discount_pct=0, region=EU, sku_total=609 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "85580", "v1": "609", "v2": "8558", "v3": "85969"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1982. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "45", "age_3": "46", "age_0": "44", "age_1": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.150", "p2": "0.050", "p0": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p3": "0.050", "p2": "0.100", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_0": "60", "answer_3": "36", "answer_1": "30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Monday, June 22, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed December 10, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 20", "date_2": "December 26", "date_3": "December 22", "date_0": "December 24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 893:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.750", "p3": "0.714", "p1": "0.250", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, March 07, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "35", "answer_1": "42", "answer_3": "49", "answer_0": "84"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_1": "28", "answer_2": "24", "answer_0": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p1": "0.625", "p2": "0.625", "p3": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $457. A fee of $46 is charged, then a deposit of $59 arrives. A second fee of $23 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "419", "answer_2": "453", "answer_0": "447", "answer_1": "476"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 08:30 to 18:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "9h15", "dur_3": "8h45", "dur_1": "9h45", "dur_2": "9h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1974. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "52", "age_0": "50", "age_1": "49", "age_2": "51"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed October 21, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 31", "date_0": "October 29", "date_1": "October 27", "date_3": "October 29"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1786 per year. It increases by 15% this year, then a $20 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2053.9", "answer_3": "2065.9", "answer_0": "2033.9", "answer_2": "1518.1"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed October 26, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 31", "date_2": "October 31", "date_1": "October 29", "date_0": "October 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 476:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.222", "p0": "0.778", "p2": "0.222", "p3": "0.750"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "36", "answer_1": "43", "answer_0": "78", "answer_3": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 424:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1245 per year. It increases by 25% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1556.25", "answer_3": "1519.25", "answer_2": "933.75", "answer_0": "1524.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1970. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "54", "age_1": "53", "age_3": "56", "age_2": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 722:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, February 05, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "36", "answer_0": "48", "answer_1": "38", "answer_3": "12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, January 27, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $711. A fee of $65 is charged, then a deposit of $80 arrives. A second fee of $32 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "697", "answer_1": "669", "answer_3": "679", "answer_0": "694"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 920:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $305 per year. It increases by 15% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "338.75", "answer_1": "350.75", "answer_3": "306.75", "answer_2": "259.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 342 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "101", "answer_3": "101", "answer_2": "108", "answer_0": "105"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.250", "p2": "0.083", "p0": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 319 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "115", "answer_1": "146", "answer_0": "126", "answer_2": "136"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1676 per year. It increases by 25% this year, then a $16 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2095", "answer_2": "1257", "answer_0": "2079", "answer_3": "2099"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = $413; priority = low.\n\nClaim to verify: \"the date is $413\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.333", "p0": "0.250", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $262 per year. It increases by 25% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "288.5", "answer_1": "327.5", "answer_3": "266.5", "answer_2": "196.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.667", "p1": "0.333", "p3": "0.600"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = $442; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 3 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "9", "answer_0": "21", "answer_2": "12", "answer_1": "15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 11:15 to 19:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "8h30", "dur_0": "8h00", "dur_2": "8h30", "dur_3": "7h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1899 per year. It increases by 15% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2141.85", "answer_1": "2183.85", "answer_3": "2121.85", "answer_2": "1614.15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.375", "p2": "0.125", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 856:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed September 09, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "September 21", "date_0": "September 23", "date_1": "September 19", "date_2": "September 21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1981. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "46", "age_1": "44", "age_3": "43", "age_0": "45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = $766; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "55", "answer_0": "104", "answer_3": "56", "answer_2": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $729. A fee of $47 is charged, then a deposit of $43 arrives. A second fee of $23 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "702", "answer_2": "690", "answer_1": "706", "answer_3": "684"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = 2026-05-08; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 08:15 to 10:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "2h00", "dur_3": "1h00", "dur_0": "1h15", "dur_2": "1h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed November 23, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 04", "date_1": "November 30", "date_0": "December 02", "date_3": "December 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, November 01, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 7 grams of salt per serving and serves 7 people. A shopper bought a 49 gram box and a 490 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1986. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "40", "age_1": "37", "age_2": "39", "age_0": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1960. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "65", "age_2": "67", "age_3": "64", "age_0": "66"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_1": "28", "answer_2": "24", "answer_0": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1960. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "62", "age_0": "64", "age_2": "65", "age_1": "63"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $346. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=563, unit_price=548, discount_pct=0, region=EU, sku_total=1111 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "308524", "v3": "309087", "v1": "1111", "v2": "30852"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p3": "0.050", "p1": "0.250", "p2": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.467", "p0": "0.533", "p1": "0.467", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = 2026-11-11; priority = high.\n\nClaim to verify: \"the owner is 2026-11-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=299, unit_price=506, discount_pct=0, region=EU, sku_total=805 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "15129", "v3": "151593", "v1": "805", "v0": "151294"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1960. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "67", "age_2": "66", "age_0": "65", "age_1": "64"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p3": "0.167", "p0": "0.500", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $726 per year. It increases by 5% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "689.7", "answer_0": "752.3", "answer_3": "768.3", "answer_1": "762.3"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=403, unit_price=505, discount_pct=0, region=EU, sku_total=908 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "20351", "v3": "203918", "v0": "203515", "v1": "908"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $26k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 299 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "57", "answer_0": "60", "answer_1": "54", "answer_2": "77"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1970. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "57", "age_3": "54", "age_0": "56", "age_1": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "30", "answer_2": "10", "answer_1": "16", "answer_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed December 07, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 17", "date_3": "December 17", "date_1": "December 15", "date_2": "December 19"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed June 15, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 24", "date_3": "June 26", "date_0": "June 26", "date_1": "June 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.222", "p3": "0.125", "p2": "0.778", "p1": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, December 26, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $639. A fee of $138 is charged, then a deposit of $30 arrives. A second fee of $69 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "477", "answer_0": "462", "answer_1": "460", "answer_2": "472"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 369 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "55", "answer_3": "68", "answer_0": "63", "answer_1": "74"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 7 people. A shopper bought a 14 gram box and a 140 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = closed; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $380. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = $849; priority = high.\n\nClaim to verify: \"the status is $849\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.800", "p3": "0.778", "p1": "0.200", "p2": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed March 11, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 20", "date_1": "March 16", "date_3": "March 18", "date_0": "March 18"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.167", "p0": "0.667", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 5 grams of salt per serving and serves 3 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = $177; priority = low.\n\nClaim to verify: \"the status is $177\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1654 per year. It increases by 15% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1405.9", "answer_3": "1837.1", "answer_1": "1902.1", "answer_0": "1870.1"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p1": "0.333", "p2": "0.333", "p3": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: status = 2026-08-23; priority = high.\n\nClaim to verify: \"the status is 2026-08-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=506, unit_price=516, discount_pct=0, region=EU, sku_total=1022 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1022", "v2": "26109", "v0": "261096", "v3": "260590"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed June 25, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 05", "date_2": "July 05", "date_1": "July 03", "date_0": "July 07"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p1": "0.250", "p3": "0.083", "p2": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = $381; priority = low.\n\nClaim to verify: \"the owner is $381\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed June 08, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 16", "date_1": "June 16", "date_0": "June 18", "date_3": "June 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1965. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "58", "age_2": "60", "age_0": "59", "age_3": "57"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1984. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "41", "age_2": "42", "age_1": "40", "age_3": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1981. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "44", "age_3": "42", "age_2": "45", "age_1": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_0": "28", "answer_1": "16", "answer_2": "14"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.333", "p0": "0.250", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=813, unit_price=202, discount_pct=0, region=EU, sku_total=1015 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "164226", "v3": "165039", "v2": "16422", "v1": "1015"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1245 per year. It increases by 15% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1436.75", "answer_1": "1431.75", "answer_2": "1058.25", "answer_0": "1399.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed May 09, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 18", "date_2": "May 20", "date_3": "May 17", "date_1": "May 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p3": "0.333", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.143", "p0": "0.250", "p1": "0.750", "p2": "0.750"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 286 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "169", "answer_3": "149", "answer_0": "160", "answer_2": "179"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 765:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "53", "age_3": "50", "age_0": "52", "age_1": "51"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1976. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "51", "age_1": "49", "age_0": "50", "age_3": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed November 08, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 14", "date_0": "November 16", "date_3": "November 16", "date_2": "November 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 200:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 16, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 29", "date_2": "May 01", "date_1": "April 25", "date_3": "April 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = $773; priority = high.\n\nClaim to verify: \"the amount is $773\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=584, unit_price=526, discount_pct=0, region=EU, sku_total=1110 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "307184", "v2": "30718", "v3": "306600", "v1": "1110"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 10 people. A shopper bought a 70 gram box and a 700 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.833", "p0": "0.667", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed May 19, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 29", "date_3": "May 31", "date_0": "June 02", "date_2": "May 31"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $331. A fee of $165 is charged, then a deposit of $46 arrives. A second fee of $82 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "139", "answer_2": "123", "answer_0": "130", "answer_3": "117"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $470; priority = low.\n\nClaim to verify: \"the owner is $470\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.250", "p2": "0.167", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, September 19, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $731 per year. It increases by 25% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "892.75", "answer_1": "913.75", "answer_0": "899.75", "answer_2": "548.25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = $781; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_0": "28", "answer_2": "16", "answer_1": "19"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1970. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "56", "age_0": "55", "age_3": "57", "age_1": "54"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.500", "p2": "0.333", "p1": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, July 25, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed March 24, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 07", "date_1": "April 03", "date_3": "April 05", "date_2": "April 09"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=689, unit_price=596, discount_pct=0, region=EU, sku_total=1285 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "41064", "v0": "410644", "v3": "411333", "v1": "1285"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.667", "p2": "0.333", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "14", "answer_1": "16", "answer_0": "28", "answer_3": "14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1966. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "60", "age_3": "58", "age_1": "59", "age_2": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $512. A fee of $40 is charged, then a deposit of $50 arrives. A second fee of $20 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "480", "answer_3": "498", "answer_1": "477", "answer_0": "502"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 238 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "63", "answer_3": "68", "answer_2": "54", "answer_0": "48"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p2": "0.083", "p3": "0.083", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "28", "answer_1": "15", "answer_0": "36", "answer_2": "8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $887. A fee of $156 is charged, then a deposit of $87 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "740", "answer_3": "718", "answer_1": "759", "answer_2": "761"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Wednesday, June 10, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $310. A fee of $107 is charged, then a deposit of $86 arrives. A second fee of $53 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "249", "answer_3": "212", "answer_0": "236", "answer_1": "245"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 8 grams of salt per serving and serves 11 people. A shopper bought a 88 gram box and a 880 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.286", "p2": "0.714", "p3": "0.167", "p1": "0.714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = 2026-08-09; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 5 grams of salt per serving and serves 12 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.150", "p0": "0.200", "p1": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = $337; priority = low.\n\nClaim to verify: \"the amount is $337\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 355 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "216", "answer_0": "210", "answer_1": "219", "answer_3": "194"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.625", "p2": "0.375", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $508 per year. It increases by 5% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "549.4", "answer_1": "533.4", "answer_0": "511.4", "answer_2": "482.6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Contract signed April 18, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 23", "date_3": "April 24", "date_2": "April 21", "date_1": "April 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.375", "p2": "0.125", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 374 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "53", "answer_0": "60", "answer_1": "45", "answer_2": "45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $875. A fee of $165 is charged, then a deposit of $82 arrives. A second fee of $82 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "690", "answer_2": "694", "answer_3": "728", "answer_0": "710"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 959:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $614 per year. It increases by 25% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "718.5", "answer_0": "738.5", "answer_1": "767.5", "answer_2": "460.5"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 192 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_2": "76", "answer_1": "98", "answer_3": "87"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $523. A fee of $156 is charged, then a deposit of $83 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "368", "answer_2": "366", "answer_1": "395", "answer_0": "372"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 08:45 to 14:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "5h30", "dur_0": "5h15", "dur_3": "4h30", "dur_2": "5h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 370 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "10", "answer_1": "28", "answer_3": "7", "answer_0": "24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=727, unit_price=526, discount_pct=0, region=EU, sku_total=1253 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1253", "v3": "381675", "v2": "38240", "v0": "382402"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 4 grams of salt per serving and serves 6 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.600", "p0": "0.400", "p1": "0.600", "p3": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = $604; priority = low.\n\nClaim to verify: \"the owner is $604\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.385", "p1": "0.385", "p0": "0.615", "p3": "0.583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed March 08, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 18", "date_2": "March 20", "date_1": "March 16", "date_3": "March 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1982. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "46", "age_0": "44", "age_1": "43", "age_2": "45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1914 per year. It increases by 5% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1939.7", "answer_0": "1972.7", "answer_1": "2009.7", "answer_2": "1818.3"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 333 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_1": "135", "answer_2": "128", "answer_3": "105"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "34", "answer_2": "30", "answer_3": "24", "answer_0": "54"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $776. A fee of $74 is charged, then a deposit of $68 arrives. A second fee of $37 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "726", "answer_1": "760", "answer_2": "728", "answer_0": "733"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1747 per year. It increases by 15% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2009.05", "answer_3": "2000.05", "answer_2": "1484.95", "answer_0": "1984.05"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p1": "0.250", "p3": "0.050", "p2": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 78 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "9", "answer_3": "32", "answer_1": "38", "answer_0": "20"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, March 04, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=519, unit_price=495, discount_pct=0, region=EU, sku_total=1014 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "256905", "v3": "256386", "v2": "25690", "v1": "1014"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = $802; priority = low.\n\nClaim to verify: \"the date is $802\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1359 per year. It increases by 15% this year, then a $49 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1513.85", "answer_3": "1536.85", "answer_2": "1155.15", "answer_1": "1562.85"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.333", "p3": "0.167", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, July 15, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "24", "answer_2": "18", "answer_3": "36", "answer_0": "54"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1793 per year. It increases by 15% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2039.95", "answer_2": "1524.05", "answer_1": "2061.95", "answer_0": "2028.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $869. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = 2026-01-09; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed November 16, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 26", "date_0": "November 30", "date_2": "November 28", "date_3": "November 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = 2026-03-26; priority = low.\n\nClaim to verify: \"the owner is 2026-03-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 5 grams of salt per serving and serves 12 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = $814; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $10k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $635. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $39k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $775. A fee of $85 is charged, then a deposit of $59 arrives. A second fee of $42 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "687", "answer_2": "686", "answer_0": "707", "answer_3": "703"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=497, unit_price=680, discount_pct=0, region=EU, sku_total=1177 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "338457", "v1": "1177", "v2": "33796", "v0": "337960"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1012 per year. It increases by 5% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "961.4", "answer_1": "1062.6", "answer_0": "1015.6", "answer_3": "1038.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1959. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "67", "age_3": "69", "age_2": "68", "age_1": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = 2026-10-12; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=781, unit_price=855, discount_pct=0, region=EU, sku_total=1636 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1636", "v0": "667755", "v2": "66775", "v3": "666974"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Sunday, May 10, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed May 08, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 11", "date_3": "May 13", "date_1": "May 11", "date_0": "May 13"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p3": "0.083", "p2": "0.083", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "40", "answer_3": "24", "answer_0": "64", "answer_1": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 11 people. A shopper bought a 99 gram box and a 990 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Wednesday, June 17, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = 2026-01-18; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $714. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed August 25, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 02", "date_2": "August 31", "date_1": "August 31", "date_3": "September 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=685, unit_price=830, discount_pct=0, region=EU, sku_total=1515 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "568550", "v3": "569235", "v1": "1515", "v2": "56855"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.455", "p1": "0.545", "p3": "0.400", "p2": "0.545"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = 2026-07-25; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "32", "answer_3": "16", "answer_1": "34", "answer_0": "48"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed February 02, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 12", "date_0": "February 12", "date_2": "February 10", "date_1": "February 10"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=437, unit_price=390, discount_pct=0, region=EU, sku_total=827 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "827", "v3": "170867", "v2": "17043", "v0": "170430"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = $947; priority = low.\n\nClaim to verify: \"the date is $947\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.800", "p0": "0.200", "p1": "0.800", "p3": "0.111"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=863, unit_price=276, discount_pct=0, region=EU, sku_total=1139 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1139", "v2": "23818", "v0": "238188", "v3": "239051"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.625", "p3": "0.286", "p1": "0.625", "p0": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 8 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = open; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $554. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 368 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "237", "answer_2": "237", "answer_3": "257", "answer_0": "240"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $461. A fee of $179 is charged, then a deposit of $25 arrives. A second fee of $89 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "199", "answer_1": "239", "answer_0": "218", "answer_2": "232"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $417. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1974. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "51", "age_1": "49", "age_0": "50", "age_3": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 818:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 529:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=420, unit_price=520, discount_pct=0, region=EU, sku_total=940 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "217980", "v0": "218400", "v1": "940", "v2": "21840"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed November 28, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 05", "date_0": "December 03", "date_1": "December 02", "date_3": "December 04"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.417", "p2": "0.583", "p1": "0.583", "p3": "0.364"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, September 01, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=453, unit_price=728, discount_pct=0, region=EU, sku_total=1181 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1181", "v2": "32978", "v0": "329784", "v3": "329331"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 10 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Saturday, June 13, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = 2026-07-17; priority = low.\n\nClaim to verify: \"the amount is 2026-07-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=497, unit_price=502, discount_pct=0, region=EU, sku_total=999 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "24949", "v1": "999", "v0": "249494", "v3": "248997"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 3 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "10", "answer_3": "15", "answer_0": "25", "answer_1": "13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 125 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "204", "answer_3": "216", "answer_1": "227", "answer_0": "224"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $644 per year. It increases by 20% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "746.8", "answer_3": "709.8", "answer_1": "772.8", "answer_2": "515.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1405 per year. It increases by 25% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1710.25", "answer_2": "1053.75", "answer_0": "1735.25", "answer_1": "1756.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed June 03, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 10", "date_3": "June 12", "date_2": "June 14", "date_0": "June 12"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 8 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1966. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "57", "age_2": "59", "age_3": "60", "age_0": "58"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = 2026-03-14; priority = high.\n\nClaim to verify: \"the status is 2026-03-14\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = $159; priority = low.\n\nClaim to verify: \"the status is $159\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1990. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "36", "age_0": "35", "age_3": "37", "age_1": "34"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = $311; priority = low.\n\nClaim to verify: \"the status is $311\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = 2026-10-15; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 9 grams of salt per serving and serves 3 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = ops; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $11k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.083", "p0": "0.167", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = $199; priority = low.\n\nClaim to verify: \"the amount is $199\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p3": "0.083", "p1": "0.250", "p0": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.667", "p3": "0.636", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "25", "answer_3": "35", "answer_1": "32", "answer_0": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_2": "18", "answer_3": "42", "answer_1": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 325:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 912:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $824. A fee of $185 is charged, then a deposit of $55 arrives. A second fee of $92 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "613", "answer_3": "611", "answer_0": "602", "answer_1": "585"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.333", "p2": "0.667", "p3": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 11, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 23", "date_3": "April 22", "date_1": "April 20", "date_2": "April 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=415, unit_price=212, discount_pct=0, region=EU, sku_total=627 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "87565", "v1": "627", "v0": "87980", "v2": "8798"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.429", "p0": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed September 24, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 03", "date_2": "October 09", "date_3": "October 05", "date_0": "October 07"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 449:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $314. A fee of $37 is charged, then a deposit of $63 arrives. A second fee of $18 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "322", "answer_3": "343", "answer_1": "313", "answer_2": "334"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 947:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 18:45 to 22:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "3h00", "dur_1": "3h45", "dur_3": "2h45", "dur_2": "3h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p1": "0.667", "p3": "0.250", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.462", "p2": "0.538", "p3": "0.417", "p1": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.364", "p0": "0.636", "p3": "0.600", "p1": "0.364"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = 2026-02-20; priority = low.\n\nClaim to verify: \"the amount is 2026-02-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, May 26, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $394 per year. It increases by 20% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "472.8", "answer_0": "429.8", "answer_3": "465.8", "answer_2": "315.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=621, unit_price=624, discount_pct=0, region=EU, sku_total=1245 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "387504", "v2": "38750", "v1": "1245", "v3": "386883"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 08:00 to 08:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h15", "dur_1": "0h15", "dur_0": "0h00", "dur_2": "0h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1980. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "45", "age_1": "43", "age_0": "44", "age_3": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.462", "p1": "0.500", "p0": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, December 09, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.111", "p0": "0.200", "p1": "0.800", "p2": "0.800"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1094 per year. It increases by 15% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1196.1", "answer_0": "1236.1", "answer_2": "929.9", "answer_1": "1258.1"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.250", "p2": "0.167", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = 2026-02-01; priority = high.\n\nClaim to verify: \"the amount is 2026-02-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p1": "0.833", "p3": "0.167", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 9 grams of salt per serving and serves 4 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=582, unit_price=109, discount_pct=0, region=EU, sku_total=691 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "6343", "v1": "691", "v0": "63438", "v3": "62856"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p3": "0.467", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Wednesday, April 15, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 393:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.375", "p1": "0.500", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $649. A fee of $99 is charged, then a deposit of $88 arrives. A second fee of $49 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "598", "answer_0": "589", "answer_1": "565", "answer_2": "562"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1984. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "44", "age_1": "41", "age_0": "42", "age_2": "43"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = 2026-08-06; priority = low.\n\nClaim to verify: \"the owner is 2026-08-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = $522; priority = high.\n\nClaim to verify: \"the owner is $522\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = 2026-06-08; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $19k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, November 21, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, January 02, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p3": "0.050", "p1": "0.250", "p0": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $608 per year. It increases by 25% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "727", "answer_0": "733", "answer_2": "456", "answer_1": "760"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 08:15 to 21:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "12h45", "dur_3": "11h45", "dur_2": "12h15", "dur_0": "12h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.800", "p3": "0.111", "p0": "0.200", "p1": "0.800"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.500", "p1": "0.667", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 254 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "107", "answer_0": "126", "answer_1": "121", "answer_3": "137"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = 2026-09-18; priority = low.\n\nClaim to verify: \"the status is 2026-09-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = $19; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1381 per year. It increases by 5% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1391.05", "answer_0": "1423.05", "answer_2": "1311.95", "answer_1": "1450.05"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1981. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "41", "age_0": "43", "age_2": "44", "age_1": "42"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p1": "0.500", "p0": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed February 03, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "February 04", "date_1": "February 06", "date_0": "February 06", "date_3": "February 08"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 238:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $464 per year. It increases by 25% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "348", "answer_3": "534", "answer_0": "559", "answer_1": "580"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1139 per year. It increases by 10% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1025.1", "answer_0": "1230.9", "answer_3": "1217.9", "answer_1": "1252.9"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1979. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "43", "age_1": "44", "age_2": "46", "age_0": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = $553; priority = low.\n\nClaim to verify: \"the amount is $553\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.533", "p1": "0.467", "p2": "0.467", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 436:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.429", "p0": "0.571", "p2": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $764. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 10:45 to 11:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "1h00", "dur_3": "0h00", "dur_0": "0h45", "dur_2": "1h00"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1974. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "51", "age_1": "49", "age_3": "52", "age_0": "50"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=639, unit_price=604, discount_pct=0, region=EU, sku_total=1243 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "38595", "v1": "1243", "v3": "385317", "v0": "385956"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=708, unit_price=197, discount_pct=0, region=EU, sku_total=905 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "13947", "v1": "905", "v3": "138768", "v0": "139476"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, July 27, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 12 people. A shopper bought a 84 gram box and a 840 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=910, unit_price=233, discount_pct=0, region=EU, sku_total=1143 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "212030", "v1": "1143", "v3": "211120", "v2": "21203"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = $880; priority = low.\n\nClaim to verify: \"the status is $880\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = $464; priority = high.\n\nClaim to verify: \"the date is $464\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = 2026-08-12; priority = low.\n\nClaim to verify: \"the amount is 2026-08-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1222 per year. It increases by 15% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1405.3", "answer_2": "1038.7", "answer_3": "1405.3", "answer_0": "1387.3"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.417", "p3": "0.083", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = 2026-09-24; priority = high.\n\nClaim to verify: \"the status is 2026-09-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed March 23, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 28", "date_3": "March 30", "date_2": "April 01", "date_0": "March 30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1970. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "54", "age_1": "53", "age_2": "55", "age_3": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 7 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "104", "answer_1": "55", "answer_3": "56", "answer_2": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $244 per year. It increases by 5% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "220.2", "answer_1": "256.2", "answer_0": "234.2", "answer_2": "231.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 337 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "116", "answer_0": "105", "answer_2": "86", "answer_1": "87"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = 2026-08-17; priority = low.\n\nClaim to verify: \"the amount is 2026-08-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 97 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "124", "answer_3": "149", "answer_0": "140", "answer_1": "138"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Thursday, March 12, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=631, unit_price=427, discount_pct=0, region=EU, sku_total=1058 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "268806", "v2": "26943", "v0": "269437", "v1": "1058"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = $458; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 511:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p0": "0.100", "p3": "0.050", "p2": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "27", "answer_1": "17", "answer_3": "15", "answer_2": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 342:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.667", "p0": "0.700", "p2": "0.300", "p1": "0.300"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1347 per year. It increases by 25% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1668.75", "answer_1": "1683.75", "answer_2": "1010.25", "answer_3": "1693.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $741. A fee of $81 is charged, then a deposit of $23 arrives. A second fee of $40 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "660", "answer_0": "643", "answer_3": "620", "answer_1": "641"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "65", "age_1": "63", "age_3": "62", "age_0": "64"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed August 19, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 23", "date_3": "August 25", "date_0": "August 25", "date_2": "August 23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = 2026-04-25; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $23k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = $11; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p3": "0.111", "p2": "0.800", "p1": "0.800"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $569. A fee of $159 is charged, then a deposit of $80 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "411", "answer_3": "436", "answer_2": "389", "answer_1": "424"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 06:30 to 08:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "1h00", "dur_0": "1h15", "dur_1": "2h00", "dur_2": "1h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "54", "answer_2": "48", "answer_3": "48", "answer_0": "96"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, February 27, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1984. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "39", "age_2": "42", "age_0": "41", "age_1": "40"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed September 18, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 27", "date_2": "September 29", "date_3": "September 29", "date_0": "October 01"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1971. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "56", "age_0": "54", "age_1": "53", "age_2": "55"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 9 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1912 per year. It increases by 10% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2103.2", "answer_3": "2062.2", "answer_0": "2091.2", "answer_2": "1720.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 389 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "135", "answer_1": "126", "answer_2": "122", "answer_0": "120"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.273", "p3": "0.700", "p1": "0.273", "p0": "0.727"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.333", "p1": "0.417", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1964. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "60", "age_3": "62", "age_1": "59", "age_2": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = $717; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = 2026-12-06; priority = low.\n\nClaim to verify: \"the status is 2026-12-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.273", "p0": "0.727", "p2": "0.273", "p3": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "19", "answer_2": "16", "answer_3": "12", "answer_0": "28"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p1": "0.500", "p3": "0.125", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $465. A fee of $56 is charged, then a deposit of $46 arrives. A second fee of $28 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "454", "answer_3": "416", "answer_0": "427", "answer_1": "449"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $532; priority = high.\n\nClaim to verify: \"the owner is $532\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=232, unit_price=185, discount_pct=0, region=EU, sku_total=417 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "43152", "v1": "417", "v0": "42920", "v2": "4292"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed January 23, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 31", "date_2": "February 02", "date_3": "February 02", "date_0": "February 04"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 331 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "160", "answer_3": "165", "answer_1": "154", "answer_2": "179"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 08:15 to 12:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "3h15", "dur_2": "3h45", "dur_0": "3h30", "dur_1": "4h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed January 15, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 21", "date_1": "January 19", "date_3": "January 21", "date_2": "January 19"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 9 grams of salt per serving and serves 4 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.333", "p1": "0.417", "p3": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=833, unit_price=465, discount_pct=0, region=EU, sku_total=1298 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "386512", "v0": "387345", "v2": "38734", "v1": "1298"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed April 25, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "May 01", "date_3": "May 03", "date_2": "May 06", "date_0": "May 04"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $317 per year. It increases by 15% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "269.45", "answer_3": "330.55", "answer_0": "316.55", "answer_1": "364.55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.714", "p1": "0.714", "p3": "0.167", "p0": "0.286"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $16k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $864. A fee of $76 is charged, then a deposit of $78 arrives. A second fee of $38 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "842", "answer_0": "828", "answer_3": "826", "answer_1": "836"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 17:00 to 22:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "4h30", "dur_1": "5h30", "dur_2": "5h30", "dur_0": "5h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed November 10, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 21", "date_1": "November 17", "date_3": "November 19", "date_0": "November 19"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = $524; priority = high.\n\nClaim to verify: \"the date is $524\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 178 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "52", "answer_2": "55", "answer_0": "36", "answer_3": "29"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 316:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1963. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "62", "age_3": "59", "age_1": "60", "age_0": "61"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=829, unit_price=865, discount_pct=0, region=EU, sku_total=1694 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1694", "v2": "71708", "v0": "717085", "v3": "717914"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $464. A fee of $167 is charged, then a deposit of $49 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "248", "answer_1": "271", "answer_0": "263", "answer_3": "275"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1972. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "55", "age_1": "53", "age_3": "56", "age_0": "54"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $8k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.333", "p1": "0.417", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 336 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "126", "answer_2": "144", "answer_1": "146", "answer_3": "137"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, April 25, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1980. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "46", "age_1": "45", "age_2": "47", "age_3": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1116 per year. It increases by 20% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "892.8", "answer_3": "1255.2", "answer_0": "1292.2", "answer_1": "1339.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 188 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "82", "answer_3": "72", "answer_0": "70", "answer_2": "50"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 343 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "55", "answer_1": "85", "answer_2": "80", "answer_0": "75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "40", "answer_3": "16", "answer_0": "56", "answer_1": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 21:00 to 22:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "1h30", "dur_3": "0h45", "dur_0": "1h00", "dur_1": "1h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = 2026-01-24; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = $655; priority = high.\n\nClaim to verify: \"the amount is $655\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=954, unit_price=543, discount_pct=0, region=EU, sku_total=1497 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "518976", "v2": "51802", "v0": "518022", "v1": "1497"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1973. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "51", "age_2": "52", "age_3": "49", "age_1": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_0": "40", "answer_3": "24", "answer_1": "22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1987. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "41", "age_0": "39", "age_1": "38", "age_2": "40"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = 2026-10-07; priority = high.\n\nClaim to verify: \"the date is 2026-10-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = $775; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 359 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "57", "answer_1": "62", "answer_0": "60", "answer_2": "42"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 8 grams of salt per serving and serves 5 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=198, unit_price=654, discount_pct=0, region=EU, sku_total=852 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "129294", "v2": "12949", "v0": "129492", "v1": "852"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1876 per year. It increases by 20% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1500.8", "answer_0": "2229.2", "answer_1": "2251.2", "answer_3": "2189.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = 2026-05-07; priority = high.\n\nClaim to verify: \"the amount is 2026-05-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 911:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p0": "0.500", "p3": "0.467"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p0": "0.333", "p3": "0.083", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = $701; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.083", "p1": "0.417", "p0": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.467", "p1": "0.467", "p3": "0.500", "p0": "0.533"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=240, unit_price=669, discount_pct=0, region=EU, sku_total=909 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "909", "v0": "160560", "v3": "160320", "v2": "16056"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.333", "p2": "0.167", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $53. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $350 per year. It increases by 25% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "451.5", "answer_0": "420.5", "answer_2": "262.5", "answer_1": "437.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $410. A fee of $78 is charged, then a deposit of $72 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "356", "answer_3": "361", "answer_1": "389", "answer_0": "365"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $867. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: date = $331; priority = low.\n\nClaim to verify: \"the date is $331\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $746. A fee of $110 is charged, then a deposit of $72 arrives. A second fee of $55 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "644", "answer_1": "670", "answer_2": "682", "answer_0": "653"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1972. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "51", "age_0": "52", "age_2": "53", "age_3": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 9 grams of salt per serving and serves 4 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "53", "age_1": "51", "age_0": "52", "age_3": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed April 02, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 05", "date_2": "April 05", "date_0": "April 07", "date_3": "April 07"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.429", "p0": "0.500", "p1": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1735 per year. It increases by 25% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2149.75", "answer_1": "2168.75", "answer_2": "1301.25", "answer_3": "2132.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "45", "answer_1": "21", "answer_2": "15", "answer_3": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_3": "40", "answer_2": "16", "answer_1": "21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p0": "0.250", "p3": "0.125", "p1": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.500", "p2": "0.167", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.333", "p0": "0.250", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = 2026-11-05; priority = low.\n\nClaim to verify: \"the amount is 2026-11-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = $104; priority = low.\n\nClaim to verify: \"the owner is $104\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 12:00 to 16:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "4h30", "dur_0": "4h15", "dur_1": "4h45", "dur_3": "3h45"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 764:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 8 grams of salt per serving and serves 9 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, August 27, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 665:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=115, unit_price=583, discount_pct=0, region=EU, sku_total=698 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "6704", "v0": "67045", "v3": "67160", "v1": "698"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = 2026-07-21; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = $741; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 269 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "16", "answer_2": "22", "answer_0": "30", "answer_3": "35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "80", "answer_1": "38", "answer_3": "48", "answer_2": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1976. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "49", "age_2": "51", "age_0": "50", "age_3": "48"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = 2026-04-24; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=716, unit_price=871, discount_pct=0, region=EU, sku_total=1587 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "62363", "v0": "623636", "v3": "624352", "v1": "1587"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Form fields: status = $671; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed June 26, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 01", "date_2": "July 05", "date_0": "July 03", "date_3": "July 03"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Sunday, April 12, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $283. A fee of $148 is charged, then a deposit of $38 arrives. A second fee of $74 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "122", "answer_0": "99", "answer_3": "93", "answer_2": "112"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Monday, October 19, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p0": "0.150", "p1": "0.200", "p3": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = $891; priority = low.\n\nClaim to verify: \"the owner is $891\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.625", "p3": "0.125", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 5 shifts this week. A client ordered 354 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "123", "answer_3": "88", "answer_2": "119", "answer_0": "105"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.125", "p2": "0.375", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1916 per year. It increases by 5% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1965.8", "answer_1": "2011.8", "answer_2": "1820.2", "answer_3": "1980.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=928, unit_price=258, discount_pct=0, region=EU, sku_total=1186 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "23942", "v0": "239424", "v3": "240352", "v1": "1186"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 73 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "55", "answer_3": "49", "answer_1": "50", "answer_0": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 696:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $278. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.556", "p3": "0.375", "p1": "0.556", "p0": "0.444"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1504 per year. It increases by 15% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1278.4", "answer_0": "1710.6", "answer_3": "1679.6", "answer_1": "1729.6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1970. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "54", "age_2": "55", "age_1": "53", "age_3": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "42", "answer_0": "84", "answer_2": "42", "answer_1": "48"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "28", "answer_0": "40", "answer_3": "16", "answer_2": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 59 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "54", "answer_2": "70", "answer_3": "68", "answer_0": "72"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_2": "16", "answer_3": "8", "answer_1": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = 2026-07-18; priority = low.\n\nClaim to verify: \"the date is 2026-07-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $4k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $480. A fee of $67 is charged, then a deposit of $37 arrives. A second fee of $33 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "437", "answer_0": "417", "answer_3": "395", "answer_2": "414"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $766 per year. It increases by 20% this year, then a $19 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "612.8", "answer_3": "915.2", "answer_1": "919.2", "answer_0": "900.2"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = 2026-10-23; priority = high.\n\nClaim to verify: \"the date is 2026-10-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.050", "p2": "0.150", "p0": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1964. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "58", "age_2": "61", "age_0": "60", "age_1": "59"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=693, unit_price=692, discount_pct=0, region=EU, sku_total=1385 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "480249", "v0": "479556", "v1": "1385", "v2": "47955"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed March 06, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 13", "date_1": "March 09", "date_0": "March 11", "date_3": "March 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed April 08, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 15", "date_1": "April 13", "date_2": "April 13", "date_0": "April 15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 08:15 to 19:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "10h30", "dur_1": "10h45", "dur_0": "10h15", "dur_3": "9h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=634, unit_price=789, discount_pct=0, region=EU, sku_total=1423 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "500226", "v3": "500860", "v1": "1423", "v2": "50022"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, December 24, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=944, unit_price=298, discount_pct=0, region=EU, sku_total=1242 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "28131", "v3": "282256", "v1": "1242", "v0": "281312"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = 2026-05-20; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=228, unit_price=911, discount_pct=0, region=EU, sku_total=1139 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1139", "v3": "207480", "v0": "207708", "v2": "20770"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p1": "0.250", "p0": "0.200", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1974. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "48", "age_1": "49", "age_0": "50", "age_2": "51"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed July 18, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "August 01", "date_0": "July 30", "date_3": "July 29", "date_1": "July 27"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = 2026-04-12; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "50", "answer_0": "64", "answer_3": "16", "answer_2": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 18:45 to 22:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "3h30", "dur_2": "3h45", "dur_3": "2h45", "dur_1": "3h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.400", "p0": "0.500", "p1": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.727", "p3": "0.200", "p0": "0.273", "p1": "0.727"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = $664; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Friday, July 10, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 08:45 to 09:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h15", "dur_0": "0h00", "dur_2": "0h15", "dur_1": "0h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p0": "0.250", "p3": "0.083", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 237 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "43", "answer_2": "38", "answer_1": "31", "answer_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=217, unit_price=682, discount_pct=0, region=EU, sku_total=899 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "14799", "v0": "147994", "v3": "147777", "v1": "899"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1982. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "43", "age_1": "42", "age_3": "41", "age_2": "44"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed November 15, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 18", "date_1": "November 18", "date_2": "November 20", "date_3": "November 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $426. A fee of $163 is charged, then a deposit of $85 arrives. A second fee of $81 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "244", "answer_3": "244", "answer_0": "267", "answer_2": "245"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.333", "p1": "0.333", "p3": "0.636"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed June 11, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 20", "date_1": "June 18", "date_2": "June 20", "date_0": "June 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 139 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "138", "answer_2": "124", "answer_0": "120", "answer_3": "107"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 392:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 11:15 to 16:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "4h45", "dur_3": "4h30", "dur_2": "5h15", "dur_1": "5h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $51. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1116 per year. It increases by 15% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "948.6", "answer_1": "1283.4", "answer_3": "1255.4", "answer_0": "1270.4"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, November 17, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $330 per year. It increases by 5% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "313.5", "answer_0": "307.5", "answer_1": "346.5", "answer_3": "286.5"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = $980; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = 2026-01-20; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 355 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "231", "answer_2": "224", "answer_3": "207", "answer_0": "216"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $243. A fee of $113 is charged, then a deposit of $68 arrives. A second fee of $56 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "114", "answer_2": "134", "answer_0": "142", "answer_3": "137"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $690. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed December 03, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 07", "date_3": "December 09", "date_1": "December 07", "date_0": "December 09"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed July 10, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 16", "date_3": "July 18", "date_2": "July 18", "date_0": "July 20"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 313 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "152", "answer_1": "142", "answer_2": "163", "answer_0": "150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, May 24, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "40", "answer_1": "44", "answer_3": "32", "answer_0": "72"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = open; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A shift runs from 06:30 to 20:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "12h45", "dur_1": "13h30", "dur_3": "12h30", "dur_2": "13h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $552. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = 2026-07-28; priority = high.\n\nClaim to verify: \"the status is 2026-07-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: status = ops; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $507. A fee of $187 is charged, then a deposit of $55 arrives. A second fee of $93 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "285", "answer_3": "306", "answer_1": "257", "answer_0": "282"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 979:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.444", "p2": "0.500", "p0": "0.500", "p1": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = 2026-10-24; priority = low.\n\nClaim to verify: \"the amount is 2026-10-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p0": "0.500", "p3": "0.125", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed October 20, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 23", "date_2": "October 21", "date_0": "October 23", "date_3": "October 25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.375", "p2": "0.250", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $27k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1973. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "55", "age_0": "53", "age_1": "52", "age_2": "54"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "15", "answer_3": "12", "answer_0": "24", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, December 27, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 143 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "18", "answer_0": "36", "answer_3": "30", "answer_2": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=856, unit_price=713, discount_pct=0, region=EU, sku_total=1569 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "61032", "v0": "610328", "v3": "609472", "v1": "1569"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $345. A fee of $117 is charged, then a deposit of $57 arrives. A second fee of $58 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "227", "answer_1": "200", "answer_2": "219", "answer_3": "212"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1982. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "45", "age_2": "44", "age_0": "43", "age_1": "42"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 655:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "6", "answer_0": "21", "answer_2": "15", "answer_1": "17"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, July 10, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "30", "answer_2": "20", "answer_0": "50", "answer_1": "26"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, June 19, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_1": "22", "answer_0": "42", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 346 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "187", "answer_0": "168", "answer_3": "178", "answer_1": "176"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $659. A fee of $70 is charged, then a deposit of $40 arrives. A second fee of $35 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "621", "answer_1": "612", "answer_3": "601", "answer_0": "594"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = 2026-11-01; priority = high.\n\nClaim to verify: \"the date is 2026-11-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.727", "p2": "0.727", "p0": "0.273", "p3": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = $105; priority = low.\n\nClaim to verify: \"the date is $105\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_3": "36", "answer_0": "66", "answer_1": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=967, unit_price=205, discount_pct=0, region=EU, sku_total=1172 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1172", "v3": "197268", "v2": "19823", "v0": "198235"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1988. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "38", "age_2": "39", "age_1": "37", "age_3": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1086 per year. It increases by 15% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1248.9", "answer_3": "1201.9", "answer_0": "1224.9", "answer_2": "923.1"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, November 16, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed September 02, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 14", "date_2": "September 12", "date_3": "September 12", "date_1": "September 10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, March 14, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed March 17, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "March 23", "date_0": "March 23", "date_1": "March 21", "date_2": "March 25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=995, unit_price=458, discount_pct=0, region=EU, sku_total=1453 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "454715", "v1": "1453", "v2": "45571", "v0": "455710"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 759:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 6 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "15", "answer_0": "27", "answer_3": "18", "answer_2": "9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.273", "p1": "0.667", "p2": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.167", "p1": "0.500", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 7 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.467", "p3": "0.429", "p2": "0.533", "p1": "0.533"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed October 15, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 22", "date_2": "October 28", "date_3": "October 24", "date_0": "October 26"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1286 per year. It increases by 10% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1157.4", "answer_3": "1357.6", "answer_1": "1414.6", "answer_0": "1373.6"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Friday, July 17, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed April 02, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "April 06", "date_0": "April 08", "date_2": "April 06", "date_3": "April 08"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Wednesday, May 20, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.417", "p3": "0.083", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 8 grams of salt per serving and serves 12 people. A shopper bought a 96 gram box and a 960 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 776:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = 2026-08-28; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 238 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "172", "answer_1": "207", "answer_0": "192", "answer_3": "210"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $583. A fee of $169 is charged, then a deposit of $23 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "325", "answer_0": "353", "answer_1": "339", "answer_3": "344"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=324, unit_price=929, discount_pct=0, region=EU, sku_total=1253 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1253", "v3": "300672", "v2": "30099", "v0": "300996"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed January 13, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 18", "date_3": "January 20", "date_0": "January 20", "date_2": "January 22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 148 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "260", "answer_1": "242", "answer_3": "228", "answer_0": "240"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $20k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = 2026-07-15; priority = high.\n\nClaim to verify: \"the status is 2026-07-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $763. A fee of $42 is charged, then a deposit of $88 arrives. A second fee of $21 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "818", "answer_0": "788", "answer_3": "805", "answer_1": "814"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = 2026-06-13; priority = high.\n\nClaim to verify: \"the status is 2026-06-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1982. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "44", "age_3": "41", "age_1": "42", "age_0": "43"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "42", "answer_0": "56", "answer_3": "14", "answer_1": "44"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=956, unit_price=398, discount_pct=0, region=EU, sku_total=1354 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "380488", "v3": "381444", "v1": "1354", "v2": "38048"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1978. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "45", "age_0": "47", "age_1": "46", "age_2": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $10k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 07:30 to 19:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "11h45", "dur_1": "12h15", "dur_2": "12h00", "dur_3": "11h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "45", "answer_2": "10", "answer_3": "35", "answer_1": "17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.500", "p3": "0.467", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "66", "age_2": "65", "age_0": "64", "age_1": "63"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.083", "p2": "0.250", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 456:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, January 15, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $282. A fee of $174 is charged, then a deposit of $45 arrives. A second fee of $87 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "66", "answer_1": "42", "answer_2": "83", "answer_3": "73"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = $63; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = $891; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed June 23, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 03", "date_1": "July 01", "date_3": "July 03", "date_2": "July 05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=860, unit_price=209, discount_pct=0, region=EU, sku_total=1069 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1069", "v3": "178880", "v0": "179740", "v2": "17974"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.429", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 4 shifts this week. A client ordered 357 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "188", "answer_1": "204", "answer_3": "208", "answer_0": "196"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 733:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = 2026-11-27; priority = low.\n\nClaim to verify: \"the owner is 2026-11-27\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=572, unit_price=750, discount_pct=0, region=EU, sku_total=1322 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1322", "v0": "429000", "v3": "428428", "v2": "42900"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1985. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "40", "age_0": "41", "age_2": "42", "age_3": "39"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = $151; priority = low.\n\nClaim to verify: \"the amount is $151\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 945:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 829:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "28", "answer_3": "20", "answer_2": "8", "answer_1": "13"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.167", "p0": "0.667", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = $407; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $35k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 10 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p1": "0.150", "p3": "0.050", "p2": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, June 26, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1960. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "63", "age_2": "65", "age_0": "64", "age_3": "62"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.444", "p2": "0.500", "p1": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.250", "p1": "0.500", "p0": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 366 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "51", "answer_3": "50", "answer_2": "45", "answer_0": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 6 grams of salt per serving and serves 10 people. A shopper bought a 60 gram box and a 600 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "40", "answer_0": "56", "answer_1": "21", "answer_2": "16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = $790; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_0": "24", "answer_2": "12", "answer_1": "16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = 2026-08-12; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 777:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1351 per year. It increases by 5% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1385.55", "answer_2": "1283.45", "answer_3": "1356.55", "answer_1": "1418.55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1968. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "55", "age_0": "57", "age_1": "56", "age_2": "58"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1778 per year. It increases by 15% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2044.7", "answer_3": "1994.7", "answer_2": "1511.3", "answer_0": "2021.7"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed January 19, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 25", "date_0": "January 23", "date_2": "January 25", "date_1": "January 23"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 4 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $36k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = 2026-05-20; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = pending; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.600", "p1": "0.400", "p2": "0.400", "p3": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 8 grams of salt per serving and serves 11 people. A shopper bought a 88 gram box and a 880 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 12 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p1": "0.500", "p3": "0.467"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 8 grams of salt per serving and serves 5 people. A shopper bought a 40 gram box and a 400 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.444", "p1": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.167", "p3": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = 2026-06-18; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $11k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = ops; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1574 per year. It increases by 5% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1495.3", "answer_0": "1619.7", "answer_1": "1652.7", "answer_3": "1630.7"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 370:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1960. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "67", "age_1": "64", "age_0": "65", "age_2": "66"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1986. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "39", "age_1": "37", "age_3": "36", "age_0": "38"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $830. A fee of $97 is charged, then a deposit of $88 arrives. A second fee of $48 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "801", "answer_3": "768", "answer_0": "773", "answer_2": "763"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $204. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "23", "answer_0": "44", "answer_2": "16", "answer_3": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $17k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed October 23, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 03", "date_1": "November 01", "date_2": "November 03", "date_0": "November 05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.200", "p2": "0.100", "p0": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.333", "p3": "0.167", "p1": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 255 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "84", "answer_2": "90", "answer_0": "72", "answer_3": "88"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.273", "p0": "0.333", "p1": "0.667", "p2": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 12 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=519, unit_price=178, discount_pct=0, region=EU, sku_total=697 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "9238", "v0": "92382", "v3": "92901", "v1": "697"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p1": "0.375", "p3": "0.125", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p3": "0.050", "p0": "0.150", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = $451; priority = low.\n\nClaim to verify: \"the date is $451\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: date = 2026-11-08; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1070 per year. It increases by 5% this year, then a $11 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1016.5", "answer_0": "1112.5", "answer_1": "1123.5", "answer_3": "1073.5"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed September 17, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 25", "date_2": "September 27", "date_1": "September 23", "date_3": "September 25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p2": "0.150", "p3": "0.050", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $995 per year. It increases by 10% this year, then a $40 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1094.5", "answer_3": "1063.5", "answer_0": "1054.5", "answer_2": "895.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1976. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "49", "age_0": "50", "age_2": "51", "age_3": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 06:00 to 16:15 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "9h15", "dur_2": "10h00", "dur_0": "9h45", "dur_1": "10h15"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.143", "p0": "0.250", "p1": "0.750", "p2": "0.750"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_3": "24", "answer_0": "36", "answer_1": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $883. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1258 per year. It increases by 5% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1320.9", "answer_2": "1195.1", "answer_0": "1293.9", "answer_3": "1287.9"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 8 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 345 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "305", "answer_0": "288", "answer_1": "291", "answer_3": "276"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1966. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "61", "age_1": "59", "age_3": "62", "age_0": "60"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1990. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "33", "age_0": "34", "age_2": "35", "age_3": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1484 per year. It increases by 5% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1515.2", "answer_2": "1409.8", "answer_1": "1558.2", "answer_0": "1525.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p2": "0.250", "p3": "0.083", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.833", "p0": "0.667", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed April 26, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 06", "date_3": "May 06", "date_1": "May 04", "date_2": "May 04"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, issue a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 2 grams of salt per serving and serves 8 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1965. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "62", "age_1": "59", "age_0": "60", "age_2": "61"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.625", "p1": "0.625", "p0": "0.375", "p3": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $428. A fee of $109 is charged, then a deposit of $74 arrives. A second fee of $54 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "331", "answer_0": "339", "answer_3": "346", "answer_1": "343"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=906, unit_price=352, discount_pct=0, region=EU, sku_total=1258 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "319818", "v1": "1258", "v0": "318912", "v2": "31891"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $32k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1730 per year. It increases by 20% this year, then a $27 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1384", "answer_3": "2066", "answer_0": "2049", "answer_1": "2076"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 558:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 6 shifts this week. A client ordered 55 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "151", "answer_0": "162", "answer_2": "164", "answer_3": "178"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed January 15, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 21", "date_1": "January 19", "date_2": "January 19", "date_3": "January 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.455", "p3": "0.500", "p0": "0.545", "p1": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=630, unit_price=218, discount_pct=0, region=EU, sku_total=848 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "137340", "v3": "137970", "v2": "13734", "v1": "848"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p3": "0.400", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed June 28, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 02", "date_3": "July 04", "date_2": "June 30", "date_0": "July 02"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.500", "p2": "0.500", "p3": "0.467"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 857:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 4 crates per hour. It operates 7 hours per shift and runs 6 shifts this week. A client ordered 308 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "168", "answer_1": "186", "answer_2": "154", "answer_3": "179"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p1": "0.150", "p2": "0.050", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.500", "p0": "0.500", "p3": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 218 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "190", "answer_2": "183", "answer_0": "175", "answer_3": "169"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 3 grams of salt per serving and serves 9 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed November 10, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 17", "date_0": "November 17", "date_2": "November 15", "date_1": "November 15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $811 per year. It increases by 25% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1013.75", "answer_0": "976.75", "answer_3": "1002.75", "answer_2": "608.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = 2026-10-10; priority = low.\n\nClaim to verify: \"the status is 2026-10-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 631:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = $233; priority = low.\n\nClaim to verify: \"the owner is $233\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p2": "0.250", "p1": "0.500", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $264. A fee of $153 is charged, then a deposit of $42 arrives. A second fee of $76 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "77", "answer_2": "104", "answer_3": "53", "answer_1": "75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = 2026-03-05; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "77", "answer_1": "35", "answer_2": "28", "answer_3": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = $808; priority = low.\n\nClaim to verify: \"the date is $808\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 612:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A shift runs from 07:00 to 08:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "0h00", "dur_1": "1h00", "dur_0": "0h30", "dur_2": "0h45"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $216. A fee of $174 is charged, then a deposit of $29 arrives. A second fee of $87 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "-19", "answer_2": "-43", "answer_3": "-10", "answer_0": "-16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.429", "p2": "0.571", "p3": "0.385", "p1": "0.571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $10k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Thursday, October 08, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=347, unit_price=301, discount_pct=0, region=EU, sku_total=648 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "104447", "v1": "648", "v3": "104100", "v2": "10444"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1967. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "59", "age_1": "58", "age_2": "60", "age_3": "61"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = 2026-04-21; priority = high.\n\nClaim to verify: \"the status is 2026-04-21\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = open; priority = high.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.600", "p0": "0.400", "p3": "0.333", "p1": "0.600"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_1": "14", "answer_0": "32", "answer_2": "8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = 2026-07-17; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1972. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "55", "age_1": "52", "age_0": "53", "age_2": "54"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 566:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=382, unit_price=351, discount_pct=0, region=EU, sku_total=733 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "13408", "v3": "133700", "v1": "733", "v0": "134082"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=797, unit_price=479, discount_pct=0, region=EU, sku_total=1276 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "380966", "v0": "381763", "v1": "1276", "v2": "38176"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $892 per year. It increases by 10% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "957.2", "answer_2": "802.8", "answer_3": "977.2", "answer_1": "981.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p1": "0.250", "p3": "0.050", "p2": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $443. A fee of $127 is charged, then a deposit of $22 arrives. A second fee of $63 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "275", "answer_2": "295", "answer_1": "301", "answer_3": "285"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=325, unit_price=240, discount_pct=0, region=EU, sku_total=565 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "78000", "v3": "78325", "v2": "7800", "v1": "565"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = 2026-08-19; priority = low.\n\nClaim to verify: \"the status is 2026-08-19\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = $579; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $909; priority = low.\n\nClaim to verify: \"the owner is $909\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.429", "p0": "0.571", "p3": "0.538", "p1": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 07:45 to 15:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "7h45", "dur_2": "7h45", "dur_0": "7h15", "dur_3": "6h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $9k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = $220; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $466. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed February 14, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 23", "date_1": "February 20", "date_2": "February 21", "date_3": "February 22"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $488. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Friday, November 13, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=157, unit_price=338, discount_pct=0, region=EU, sku_total=495 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "53223", "v0": "53066", "v1": "495", "v2": "5306"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = 2026-12-02; priority = low.\n\nClaim to verify: \"the date is 2026-12-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = 2026-06-17; priority = high.\n\nClaim to verify: \"the amount is 2026-06-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 3 grams of salt per serving and serves 12 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = $569; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = $792; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = $472; priority = low.\n\nClaim to verify: \"the status is $472\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = $832; priority = high.\n\nClaim to verify: \"the status is $832\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1936 per year. It increases by 10% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2065.6", "answer_2": "1742.4", "answer_0": "2090.6", "answer_1": "2129.6"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.150", "p0": "0.100", "p2": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed August 10, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 16", "date_0": "August 18", "date_2": "August 16", "date_3": "August 18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $335. A fee of $118 is charged, then a deposit of $35 arrives. A second fee of $59 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "193", "answer_1": "208", "answer_3": "173", "answer_2": "167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=182, unit_price=904, discount_pct=0, region=EU, sku_total=1086 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "16452", "v1": "1086", "v3": "164346", "v0": "164528"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.083", "p1": "0.333", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = $945; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $798. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "14", "answer_0": "28", "answer_1": "16", "answer_3": "14"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $505. A fee of $108 is charged, then a deposit of $36 arrives. A second fee of $54 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "366", "answer_1": "374", "answer_0": "379", "answer_2": "401"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=679, unit_price=678, discount_pct=0, region=EU, sku_total=1357 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "46036", "v0": "460362", "v1": "1357", "v3": "461041"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 859:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.462", "p2": "0.462", "p3": "0.500", "p0": "0.538"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $28k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 377:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $662 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "663.1", "answer_2": "628.9", "answer_1": "695.1", "answer_3": "700.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 894:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $436 per year. It increases by 10% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "429.6", "answer_1": "479.6", "answer_2": "392.4", "answer_3": "460.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "27", "answer_2": "24", "answer_3": "18", "answer_0": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1978. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "45", "age_3": "44", "age_2": "47", "age_0": "46"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 11 people. A shopper bought a 88 gram box and a 880 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $604. A fee of $166 is charged, then a deposit of $65 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "428", "answer_2": "429", "answer_3": "390", "answer_0": "420"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, December 08, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $242 per year. It increases by 10% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "217.8", "answer_1": "266.2", "answer_3": "209.2", "answer_0": "225.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.800", "p0": "0.200", "p3": "0.111", "p2": "0.800"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1983. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "41", "age_2": "44", "age_1": "42", "age_0": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $21k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.125", "p1": "0.500", "p0": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 6 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $656. A fee of $65 is charged, then a deposit of $24 arrives. A second fee of $32 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "598", "answer_1": "566", "answer_2": "594", "answer_0": "583"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 111 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "13", "answer_0": "28", "answer_2": "15", "answer_1": "48"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 8 grams of salt per serving and serves 4 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 6 grams of salt per serving and serves 12 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = 2026-03-06; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $854. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed October 16, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 23", "date_2": "October 25", "date_0": "October 23", "date_1": "October 21"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1975. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "50", "age_3": "52", "age_2": "51", "age_1": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 6 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 195 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "78", "answer_1": "83", "answer_0": "72", "answer_3": "79"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, March 17, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $30k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "50", "answer_3": "30", "answer_2": "20", "answer_1": "26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.500", "p0": "0.500", "p3": "0.462"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "14", "answer_2": "21", "answer_1": "23", "answer_0": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $389. A fee of $194 is charged, then a deposit of $54 arrives. A second fee of $97 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "177", "answer_3": "157", "answer_1": "135", "answer_0": "152"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = 2026-08-16; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed July 02, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 08", "date_2": "July 10", "date_3": "July 08", "date_1": "July 06"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1976. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "47", "age_0": "48", "age_2": "49", "age_3": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $71. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1041 per year. It increases by 20% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1251.2", "answer_1": "1249.2", "answer_2": "832.8", "answer_0": "1225.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $32k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = $338; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 07, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 18", "date_1": "April 16", "date_0": "April 20", "date_2": "April 22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 09:30 to 19:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "9h30", "dur_1": "10h00", "dur_2": "9h45", "dur_3": "9h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed January 06, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 12", "date_1": "January 10", "date_3": "January 12", "date_2": "January 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1679 per year. It increases by 5% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1762.95", "answer_2": "1595.05", "answer_0": "1744.95", "answer_3": "1760.95"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed February 07, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 17", "date_0": "February 18", "date_1": "February 15", "date_2": "February 20"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1970. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "56", "age_0": "55", "age_3": "57", "age_1": "54"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Sunday, May 24, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p2": "0.600", "p1": "0.600", "p0": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed August 21, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 03", "date_1": "August 30", "date_2": "September 01", "date_3": "September 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = open; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1313 per year. It increases by 15% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1116.05", "answer_1": "1509.95", "answer_0": "1468.95", "answer_3": "1458.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $890 per year. It increases by 20% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "994", "answer_2": "712", "answer_0": "1026", "answer_1": "1068"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p0": "0.250", "p3": "0.125", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 12:00 to 12:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "0h45", "dur_0": "0h15", "dur_3": "-1h45", "dur_2": "0h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1970. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "54", "age_3": "57", "age_0": "55", "age_2": "56"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed February 06, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 11", "date_2": "February 09", "date_1": "February 09", "date_3": "February 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 243:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $546. A fee of $59 is charged, then a deposit of $28 arrives. A second fee of $29 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "486", "answer_3": "510", "answer_1": "480", "answer_2": "484"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = team-b; priority = low.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $254 per year. It increases by 15% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "292.1", "answer_0": "278.1", "answer_3": "269.1", "answer_2": "215.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $32k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = $473; priority = high.\n\nClaim to verify: \"the owner is $473\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed March 04, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 18", "date_1": "March 14", "date_2": "March 16", "date_3": "March 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 480:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, March 10, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Friday, December 04, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $40k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.375", "p3": "0.125", "p1": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.125", "p1": "0.500", "p0": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "22", "answer_0": "50", "answer_2": "15", "answer_3": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1962. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "63", "age_3": "62", "age_0": "64", "age_2": "65"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = 2026-03-13; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $807. A fee of $38 is charged, then a deposit of $40 arrives. A second fee of $19 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "814", "answer_0": "790", "answer_1": "780", "answer_2": "783"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_3": "12", "answer_0": "42", "answer_1": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 248:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed November 09, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 23", "date_3": "November 21", "date_1": "November 19", "date_2": "November 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "28", "answer_0": "49", "answer_1": "31", "answer_3": "21"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Friday, April 24, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1802 per year. It increases by 25% this year, then a $15 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2200.5", "answer_0": "2237.5", "answer_1": "2252.5", "answer_2": "1351.5"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.333", "p1": "0.333", "p3": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=438, unit_price=371, discount_pct=0, region=EU, sku_total=809 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "162498", "v2": "16249", "v1": "809", "v3": "162936"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = 2026-01-25; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $23k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=430, unit_price=653, discount_pct=0, region=EU, sku_total=1083 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "280790", "v3": "281220", "v2": "28079", "v1": "1083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.375", "p0": "0.250", "p2": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p2": "0.083", "p3": "0.083", "p0": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $894. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $784. A fee of $159 is charged, then a deposit of $53 arrives. A second fee of $79 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "623", "answer_3": "596", "answer_0": "599", "answer_1": "603"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "26", "answer_3": "12", "answer_0": "36", "answer_2": "24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = $687; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 294 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "27", "answer_2": "12", "answer_0": "20", "answer_1": "17"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $34k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=829, unit_price=176, discount_pct=0, region=EU, sku_total=1005 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "145904", "v1": "1005", "v2": "14590", "v3": "145075"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.167", "p1": "0.250", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1965 per year. It increases by 20% this year, then a $18 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1572", "answer_0": "2340", "answer_1": "2358", "answer_3": "2325"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 586:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1964. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "62", "age_0": "61", "age_3": "59", "age_1": "60"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $849. A fee of $113 is charged, then a deposit of $47 arrives. A second fee of $56 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "757", "answer_3": "746", "answer_0": "727", "answer_2": "702"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p3": "0.250", "p0": "0.400", "p2": "0.600"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1904 per year. It increases by 10% this year, then a $13 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1713.6", "answer_3": "2070.4", "answer_0": "2081.4", "answer_1": "2094.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = 2026-08-18; priority = low.\n\nClaim to verify: \"the status is 2026-08-18\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 276:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.600", "p0": "0.667", "p2": "0.333", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1979. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "47", "age_1": "46", "age_2": "48", "age_3": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=357, unit_price=227, discount_pct=0, region=EU, sku_total=584 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "8103", "v3": "80682", "v0": "81039", "v1": "584"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_3": "16", "answer_2": "24", "answer_1": "26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 397:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1968. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_3": "56", "age_2": "59", "age_1": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1972. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "54", "age_0": "53", "age_1": "52", "age_3": "51"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 210:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.400", "p0": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1961. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "65", "age_1": "64", "age_3": "67", "age_2": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, August 05, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.100", "p2": "0.050", "p1": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.833", "p0": "0.667", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1548 per year. It increases by 15% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1780.2", "answer_2": "1315.8", "answer_3": "1784.2", "answer_0": "1744.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.417", "p1": "0.417", "p3": "0.545", "p0": "0.583"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed July 09, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 17", "date_1": "July 13", "date_3": "July 15", "date_0": "July 15"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "19", "answer_3": "24", "answer_2": "16", "answer_0": "40"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 12 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $264 per year. It increases by 25% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "269", "answer_2": "198", "answer_0": "296", "answer_1": "330"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=958, unit_price=387, discount_pct=0, region=EU, sku_total=1345 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "370746", "v3": "369788", "v1": "1345", "v2": "37074"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 155 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "27", "answer_2": "19", "answer_1": "36", "answer_3": "21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=209, unit_price=990, discount_pct=0, region=EU, sku_total=1199 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1199", "v3": "206701", "v0": "206910", "v2": "20691"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "36", "answer_3": "32", "answer_2": "32", "answer_0": "64"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed October 25, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 03", "date_2": "November 01", "date_3": "November 03", "date_1": "November 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1963. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "59", "age_0": "61", "age_1": "60", "age_2": "62"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=687, unit_price=855, discount_pct=0, region=EU, sku_total=1542 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "586698", "v2": "58738", "v0": "587385", "v1": "1542"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "48", "answer_3": "16", "answer_1": "34", "answer_2": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 298 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "197", "answer_0": "192", "answer_3": "212", "answer_2": "206"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=535, unit_price=381, discount_pct=0, region=EU, sku_total=916 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "203835", "v2": "20383", "v1": "916", "v3": "203300"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.750", "p1": "0.750", "p0": "0.250", "p3": "0.143"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.364", "p3": "0.600", "p1": "0.364", "p0": "0.636"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.600", "p1": "0.400", "p3": "0.500", "p2": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, November 06, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $35k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = $104; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed October 18, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "October 28", "date_2": "October 30", "date_1": "October 26", "date_3": "October 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, July 26, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 12 people. A shopper bought a 96 gram box and a 960 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = open; priority = low.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1963. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "62", "age_1": "61", "age_2": "63", "age_3": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1976 per year. It increases by 25% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2412", "answer_1": "2470", "answer_2": "1482", "answer_0": "2428"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 4 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "28", "answer_0": "63", "answer_1": "39", "answer_2": "35"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = $220; priority = low.\n\nClaim to verify: \"the owner is $220\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 250:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = 2026-02-27; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, August 26, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=100, unit_price=585, discount_pct=0, region=EU, sku_total=685 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "685", "v0": "58500", "v2": "5850", "v3": "58600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: date = $111; priority = high.\n\nClaim to verify: \"the date is $111\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = 2026-11-28; priority = high.\n\nClaim to verify: \"the owner is 2026-11-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $5k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1967. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "59", "age_3": "57", "age_2": "60", "age_1": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 260 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "193", "answer_0": "180", "answer_3": "167", "answer_1": "198"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1959. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "67", "age_2": "68", "age_3": "65", "age_1": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $512 per year. It increases by 5% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "537.6", "answer_3": "528.6", "answer_0": "520.6", "answer_2": "486.4"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $460. A fee of $115 is charged, then a deposit of $38 arrives. A second fee of $57 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "311", "answer_3": "328", "answer_0": "326", "answer_2": "354"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = 2026-09-07; priority = low.\n\nClaim to verify: \"the amount is 2026-09-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: status = $616; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 7 grams of salt per serving and serves 7 people. A shopper bought a 49 gram box and a 490 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.150", "p1": "0.250", "p0": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A shift runs from 11:45 to 21:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "9h30", "dur_3": "9h00", "dur_1": "10h00", "dur_2": "9h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $537. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 156 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "209", "answer_2": "234", "answer_0": "224", "answer_3": "232"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = $440; priority = low.\n\nClaim to verify: \"the owner is $440\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=670, unit_price=302, discount_pct=0, region=EU, sku_total=972 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "201670", "v1": "972", "v0": "202340", "v2": "20234"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Record fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.333", "p0": "0.250", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p2": "0.667", "p1": "0.667", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "30", "answer_0": "55", "answer_1": "35", "answer_3": "25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1979. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "49", "age_0": "47", "age_1": "46", "age_2": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=338, unit_price=465, discount_pct=0, region=EU, sku_total=803 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "157170", "v3": "157508", "v1": "803", "v2": "15717"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1960. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "65", "age_1": "63", "age_0": "64", "age_3": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Saturday, December 26, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $342. A fee of $185 is charged, then a deposit of $63 arrives. A second fee of $92 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "142", "answer_2": "144", "answer_1": "99", "answer_0": "128"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p1": "0.250", "p2": "0.083", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 3 grams of salt per serving and serves 5 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, October 10, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed October 16, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "October 26", "date_3": "October 24", "date_1": "October 22", "date_2": "October 24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1096 per year. It increases by 25% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1370", "answer_0": "1324", "answer_2": "822", "answer_3": "1314"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = $455; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_1": "12", "answer_3": "16", "answer_2": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $250. A fee of $173 is charged, then a deposit of $27 arrives. A second fee of $86 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "18", "answer_1": "10", "answer_3": "26", "answer_2": "1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1970. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "53", "age_3": "52", "age_0": "54", "age_2": "55"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.417", "p3": "0.083", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 5 grams of salt per serving and serves 11 people. A shopper bought a 55 gram box and a 550 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p3": "0.083", "p1": "0.250", "p0": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = 2026-01-18; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "18", "answer_1": "15", "answer_2": "12", "answer_0": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 723:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = 2026-09-16; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: amount = 2026-04-17; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed September 02, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 11", "date_0": "September 09", "date_1": "September 07", "date_3": "September 09"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, April 12, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_2": "30", "answer_1": "32", "answer_0": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = 2026-10-09; priority = high.\n\nClaim to verify: \"the amount is 2026-10-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = 2026-11-05; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $628. A fee of $132 is charged, then a deposit of $56 arrives. A second fee of $66 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "469", "answer_1": "475", "answer_0": "486", "answer_2": "515"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 5 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p3": "0.083", "p0": "0.250", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p1": "0.625", "p3": "0.125", "p0": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $316. A fee of $190 is charged, then a deposit of $45 arrives. A second fee of $95 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "87", "answer_2": "72", "answer_0": "76", "answer_1": "86"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = closed; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1967. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "56", "age_3": "55", "age_2": "58", "age_0": "57"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $791. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = $275; priority = low.\n\nClaim to verify: \"the owner is $275\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = $637; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.400", "p0": "0.500", "p2": "0.500", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p1": "0.333", "p3": "0.636", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, December 25, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=130, unit_price=719, discount_pct=0, region=EU, sku_total=849 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "93600", "v2": "9347", "v0": "93470", "v1": "849"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $244. A fee of $31 is charged, then a deposit of $62 arrives. A second fee of $15 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "234", "answer_3": "265", "answer_1": "231", "answer_0": "260"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p1": "0.500", "p2": "0.250", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = $511; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 246:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 8 grams of salt per serving and serves 7 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.250", "p3": "0.083", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $20k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = $598; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 17:00 to 18:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "0h30", "dur_1": "1h30", "dur_0": "0h45", "dur_2": "1h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $811 per year. It increases by 25% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "977.75", "answer_2": "608.25", "answer_3": "941.75", "answer_1": "1013.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Saturday, April 11, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = 2026-07-06; priority = low.\n\nClaim to verify: \"the amount is 2026-07-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p3": "0.455", "p1": "0.500", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed December 16, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 24", "date_0": "December 24", "date_1": "December 22", "date_2": "December 26"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 171 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "189", "answer_2": "190", "answer_3": "196", "answer_0": "200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = $613; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1978. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "50", "age_1": "47", "age_0": "48", "age_2": "49"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_3": "42", "answer_0": "66", "answer_1": "31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p3": "0.636", "p0": "0.667", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.250", "p0": "0.375", "p3": "0.125"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1038 per year. It increases by 10% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1109.8", "answer_3": "1135.8", "answer_1": "1141.8", "answer_2": "934.2"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = 2026-02-06; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 332 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_2": "62", "answer_1": "76", "answer_3": "87"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $625. A fee of $71 is charged, then a deposit of $36 arrives. A second fee of $35 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "555", "answer_2": "528", "answer_1": "553", "answer_3": "533"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "84", "answer_2": "42", "answer_1": "48", "answer_3": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 412:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 936:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.455", "p0": "0.500", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: amount = 2026-05-07; priority = low.\n\nClaim to verify: \"the amount is 2026-05-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p0": "0.333", "p1": "0.667", "p3": "0.200"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = 2026-12-17; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 5 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 18 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $726. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 879:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = 2026-03-05; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1170 per year. It increases by 5% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1111.5", "answer_0": "1182.5", "answer_3": "1187.5", "answer_1": "1228.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 114 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_0": "28", "answer_1": "20", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, August 03, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "20", "answer_0": "24", "answer_2": "18", "answer_3": "6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $249. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1958. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "69", "age_3": "66", "age_0": "68", "age_1": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p3": "0.050", "p1": "0.250", "p0": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.636", "p3": "0.300", "p1": "0.636", "p0": "0.364"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed December 09, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 18", "date_2": "December 20", "date_0": "December 22", "date_3": "December 20"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $12k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 4 grams of salt per serving and serves 6 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "28", "answer_2": "24", "answer_3": "16", "answer_0": "40"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 4 grams of salt per serving and serves 11 people. A shopper bought a 44 gram box and a 440 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $412. A fee of $30 is charged, then a deposit of $87 arrives. A second fee of $15 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "477", "answer_3": "429", "answer_2": "483", "answer_0": "454"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p0": "0.250", "p3": "0.125", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $711. A fee of $70 is charged, then a deposit of $26 arrives. A second fee of $35 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "612", "answer_1": "608", "answer_3": "629", "answer_0": "632"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.778", "p0": "0.222", "p1": "0.778"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 9 people. A shopper bought a 27 gram box and a 270 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=853, unit_price=833, discount_pct=0, region=EU, sku_total=1686 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "710549", "v1": "1686", "v3": "711402", "v2": "71054"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.625", "p2": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p1": "0.333", "p2": "0.333", "p3": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 602:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1260 per year. It increases by 20% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1008", "answer_3": "1466", "answer_0": "1479", "answer_1": "1512"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $684. A fee of $128 is charged, then a deposit of $21 arrives. A second fee of $64 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "535", "answer_3": "527", "answer_0": "513", "answer_2": "536"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, July 25, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1751 per year. It increases by 20% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1400.8", "answer_3": "2098.2", "answer_1": "2101.2", "answer_0": "2066.2"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed November 04, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 09", "date_3": "November 09", "date_1": "November 07", "date_2": "November 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Wednesday, June 17, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = $82; priority = high.\n\nClaim to verify: \"the amount is $82\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $618. A fee of $154 is charged, then a deposit of $20 arrives. A second fee of $77 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "384", "answer_0": "407", "answer_1": "403", "answer_3": "409"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.100", "p0": "0.150", "p1": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $388. A fee of $186 is charged, then a deposit of $35 arrives. A second fee of $93 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "144", "answer_3": "171", "answer_1": "126", "answer_2": "171"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $893. A fee of $191 is charged, then a deposit of $81 arrives. A second fee of $95 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "702", "answer_2": "684", "answer_1": "694", "answer_0": "688"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = 2026-05-03; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "8", "answer_0": "16", "answer_1": "10", "answer_3": "8"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.545", "p2": "0.545", "p3": "0.400", "p0": "0.455"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.714", "p0": "0.286", "p1": "0.714"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed June 26, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 02", "date_3": "July 02", "date_1": "June 30", "date_2": "July 04"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=555, unit_price=437, discount_pct=0, region=EU, sku_total=992 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "992", "v0": "242535", "v3": "241980", "v2": "24253"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 275:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.500", "p0": "0.500", "p3": "0.467"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 383 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "47", "answer_3": "57", "answer_2": "75", "answer_0": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed April 04, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 10", "date_2": "April 12", "date_1": "April 09", "date_3": "April 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.462", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $6k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 5 grams of salt per serving and serves 9 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $860. A fee of $31 is charged, then a deposit of $48 arrives. A second fee of $15 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "862", "answer_1": "833", "answer_2": "865", "answer_3": "855"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1986. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "42", "age_1": "39", "age_0": "40", "age_2": "41"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed August 13, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 22", "date_2": "August 22", "date_0": "August 24", "date_1": "August 20"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p1": "0.150", "p2": "0.050", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 462:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed February 14, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 26", "date_0": "February 27", "date_2": "March 01", "date_1": "February 24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 203 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "194", "answer_3": "169", "answer_0": "180", "answer_1": "193"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "30", "answer_1": "20", "answer_3": "15", "answer_2": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.462", "p2": "0.500", "p1": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = $772; priority = high.\n\nClaim to verify: \"the date is $772\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_1": "28", "answer_3": "32", "answer_2": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = 2026-08-03; priority = high.\n\nClaim to verify: \"the status is 2026-08-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 7 grams of salt per serving and serves 4 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: status = 2026-11-09; priority = low.\n\nClaim to verify: \"the status is 2026-11-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = 2026-10-05; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=643, unit_price=127, discount_pct=0, region=EU, sku_total=770 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "770", "v3": "81018", "v2": "8166", "v0": "81661"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $16k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 500:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 59 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "56", "answer_2": "36", "answer_3": "53", "answer_0": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.538", "p1": "0.462", "p3": "0.500", "p2": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $11k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 937:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = 2026-09-27; priority = high.\n\nClaim to verify: \"the status is 2026-09-27\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p0": "0.250", "p1": "0.375", "p3": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = open; priority = high.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: amount = 2026-12-07; priority = low.\n\nClaim to verify: \"the amount is 2026-12-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1812 per year. It increases by 10% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1993.2", "answer_0": "1981.2", "answer_3": "1992.2", "answer_2": "1630.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $249. A fee of $168 is charged, then a deposit of $68 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "65", "answer_2": "77", "answer_1": "92", "answer_3": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=132, unit_price=283, discount_pct=0, region=EU, sku_total=415 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "3735", "v0": "37356", "v1": "415", "v3": "37224"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.667", "p3": "0.250", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $866. A fee of $66 is charged, then a deposit of $74 arrives. A second fee of $33 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "857", "answer_0": "841", "answer_2": "867", "answer_3": "835"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=318, unit_price=953, discount_pct=0, region=EU, sku_total=1271 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "303372", "v1": "1271", "v2": "30305", "v0": "303054"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 4 grams of salt per serving and serves 7 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 235 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "92", "answer_1": "64", "answer_0": "84", "answer_2": "94"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $38k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, January 18, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Monday, March 16, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.273", "p3": "0.700", "p0": "0.727", "p2": "0.273"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 11 people. A shopper bought a 44 gram box and a 440 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $583. A fee of $164 is charged, then a deposit of $28 arrives. A second fee of $82 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "365", "answer_3": "385", "answer_1": "348", "answer_2": "361"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 376 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "44", "answer_0": "60", "answer_3": "71", "answer_2": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = 2026-02-02; priority = low.\n\nClaim to verify: \"the owner is 2026-02-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $305. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = $654; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $459 per year. It increases by 20% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "538.8", "answer_2": "367.2", "answer_3": "504.8", "answer_1": "550.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 12 people. A shopper bought a 108 gram box and a 1080 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, November 11, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $386. A fee of $195 is charged, then a deposit of $57 arrives. A second fee of $97 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "151", "answer_2": "129", "answer_3": "161", "answer_1": "133"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.400", "p1": "0.400", "p0": "0.600", "p3": "0.556"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1966. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "59", "age_3": "57", "age_2": "60", "age_1": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = $439; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed December 21, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 01", "date_1": "December 30", "date_3": "January 01", "date_2": "December 30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, August 28, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p1": "0.625", "p3": "0.286", "p2": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed July 17, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 25", "date_2": "July 29", "date_1": "July 23", "date_0": "July 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed June 12, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 19", "date_2": "June 21", "date_3": "June 21", "date_0": "June 23"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=727, unit_price=760, discount_pct=0, region=EU, sku_total=1487 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "551793", "v2": "55252", "v1": "1487", "v0": "552520"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = $81; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 184 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "51", "answer_0": "60", "answer_2": "78", "answer_3": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=733, unit_price=298, discount_pct=0, region=EU, sku_total=1031 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1031", "v3": "219167", "v2": "21843", "v0": "218434"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = $997; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 2 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_3": "10", "answer_1": "32", "answer_2": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: date = pending; priority = high.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 309:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1986. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "40", "age_3": "41", "age_0": "39", "age_1": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 247 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "144", "answer_2": "124", "answer_3": "138", "answer_1": "156"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.571", "p2": "0.375", "p1": "0.375", "p0": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.273", "p1": "0.727", "p3": "0.200", "p2": "0.727"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 07:45 to 21:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "12h30", "dur_1": "13h15", "dur_3": "12h15", "dur_2": "13h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed March 25, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 06", "date_1": "April 04", "date_0": "April 08", "date_2": "April 06"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $509. A fee of $33 is charged, then a deposit of $66 arrives. A second fee of $16 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "515", "answer_2": "536", "answer_0": "526", "answer_3": "512"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=450, unit_price=326, discount_pct=0, region=EU, sku_total=776 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "14670", "v0": "146700", "v3": "146250", "v1": "776"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.167", "p0": "0.333", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed March 03, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "March 12", "date_3": "March 10", "date_0": "March 10", "date_1": "March 08"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.083", "p1": "0.333", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.700", "p3": "0.222", "p1": "0.700", "p0": "0.300"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.625", "p3": "0.571", "p1": "0.375", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1987. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "37", "age_1": "36", "age_3": "39", "age_2": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 979:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.286", "p1": "0.714", "p2": "0.714"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1975. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "51", "age_0": "49", "age_2": "50", "age_1": "48"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed June 20, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "June 30", "date_0": "July 03", "date_3": "July 02", "date_2": "July 05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1978. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "49", "age_1": "47", "age_3": "50", "age_0": "48"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1498 per year. It increases by 15% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1722.7", "answer_0": "1705.7", "answer_2": "1273.3", "answer_3": "1679.7"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "40", "answer_0": "64", "answer_1": "43", "answer_3": "24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = 2026-05-06; priority = high.\n\nClaim to verify: \"the owner is 2026-05-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed September 14, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 23", "date_3": "September 25", "date_0": "September 25", "date_2": "September 27"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 275 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "229", "answer_2": "213", "answer_0": "210", "answer_1": "192"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.700", "p3": "0.667", "p2": "0.300", "p1": "0.300"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.571", "p0": "0.429", "p1": "0.571", "p3": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = $127; priority = low.\n\nClaim to verify: \"the date is $127\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $772 per year. It increases by 10% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "864.2", "answer_2": "694.8", "answer_0": "825.2", "answer_1": "849.2"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.455", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 8 crates per hour. It operates 4 hours per shift and runs 4 shifts this week. A client ordered 208 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "109", "answer_1": "134", "answer_2": "120", "answer_0": "128"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Tuesday, August 25, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 556:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 7 grams of salt per serving and serves 4 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 6 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $546 per year. It increases by 10% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "600.6", "answer_3": "614.6", "answer_0": "579.6", "answer_2": "491.4"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = $579; priority = low.\n\nClaim to verify: \"the status is $579\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 751:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_1": "19", "answer_2": "16", "answer_0": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $819. A fee of $104 is charged, then a deposit of $20 arrives. A second fee of $52 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "683", "answer_2": "654", "answer_1": "685", "answer_3": "703"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p1": "0.375", "p0": "0.250", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = $175; priority = low.\n\nClaim to verify: \"the owner is $175\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.571", "p2": "0.571", "p3": "0.385", "p0": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 688:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, September 14, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, May 10, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1668 per year. It increases by 15% this year, then a $49 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1417.8", "answer_1": "1918.2", "answer_3": "1853.2", "answer_0": "1869.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $860. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 810:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 8 grams of salt per serving and serves 3 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p0": "0.100", "p3": "0.050", "p1": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "30", "answer_2": "24", "answer_0": "60", "answer_3": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "32", "answer_0": "60", "answer_2": "25", "answer_3": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Thursday, August 06, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: status = $344; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Friday, April 03, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 821:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = 2026-09-04; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: status = 2026-05-20; priority = high.\n\nClaim to verify: \"the status is 2026-05-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 665:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = open; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 382:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 5 people. A shopper bought a 45 gram box and a 450 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed September 14, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 18", "date_3": "September 20", "date_0": "September 18", "date_2": "September 20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 381:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=714, unit_price=119, discount_pct=0, region=EU, sku_total=833 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "8496", "v3": "84252", "v1": "833", "v0": "84966"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: date = pending; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.625", "p2": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = ops; priority = low.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $351. A fee of $173 is charged, then a deposit of $55 arrives. A second fee of $86 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "133", "answer_0": "147", "answer_1": "121", "answer_3": "176"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.364", "p2": "0.364", "p3": "0.600", "p0": "0.636"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Form fields: status = $194; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = $61; priority = low.\n\nClaim to verify: \"the amount is $61\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "38", "answer_3": "12", "answer_0": "48", "answer_2": "36"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = open; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Monday, October 19, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 197 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_2": "139", "answer_1": "140", "answer_3": "140"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.727", "p0": "0.273", "p3": "0.200", "p2": "0.727"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: date = closed; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 23 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed May 06, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 11", "date_1": "May 09", "date_3": "May 11", "date_2": "May 09"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 07:45 to 09:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "1h00", "dur_2": "1h45", "dur_0": "1h30", "dur_1": "2h00"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p1": "0.250", "p0": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.583", "p2": "0.583", "p3": "0.364", "p0": "0.417"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1969. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "57", "age_2": "58", "age_1": "56", "age_3": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $823. A fee of $137 is charged, then a deposit of $90 arrives. A second fee of $68 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "700", "answer_2": "712", "answer_0": "708", "answer_1": "715"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=990, unit_price=755, discount_pct=0, region=EU, sku_total=1745 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "74745", "v3": "748440", "v0": "747450", "v1": "1745"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Friday, March 13, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = $732; priority = low.\n\nClaim to verify: \"the amount is $732\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $643. A fee of $153 is charged, then a deposit of $65 arrives. A second fee of $76 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "475", "answer_2": "462", "answer_0": "479", "answer_3": "483"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = 2026-05-06; priority = high.\n\nClaim to verify: \"the owner is 2026-05-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1964. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "61", "age_1": "60", "age_3": "63", "age_2": "62"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 365 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "75", "answer_0": "90", "answer_3": "95", "answer_2": "73"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 160 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_2": "52", "answer_1": "23", "answer_0": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = 2026-06-08; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Monday, March 16, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_0": "24", "answer_3": "12", "answer_1": "16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 390 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "36", "answer_3": "41", "answer_0": "30", "answer_1": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Saturday, October 24, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $496. A fee of $86 is charged, then a deposit of $32 arrives. A second fee of $43 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "378", "answer_2": "377", "answer_0": "399", "answer_3": "420"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1970. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "57", "age_2": "56", "age_1": "54", "age_0": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = 2026-06-18; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $510 per year. It increases by 5% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "535.5", "answer_3": "475.5", "answer_2": "484.5", "answer_0": "500.5"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = $340; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 4 grams of salt per serving and serves 12 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Friday, February 06, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = 2026-12-01; priority = high.\n\nClaim to verify: \"the amount is 2026-12-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = $34; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1723 per year. It increases by 10% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1550.7", "answer_3": "1880.3", "answer_1": "1895.3", "answer_0": "1874.3"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 393:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1433 per year. It increases by 15% this year, then a $26 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1621.95", "answer_3": "1638.95", "answer_1": "1647.95", "answer_2": "1218.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $282 per year. It increases by 10% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "253.8", "answer_0": "285.2", "answer_1": "310.2", "answer_3": "302.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $33k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $210. A fee of $156 is charged, then a deposit of $85 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "61", "answer_2": "68", "answer_3": "83", "answer_1": "77"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = $857; priority = low.\n\nClaim to verify: \"the status is $857\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.429", "p1": "0.429", "p0": "0.571"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: status = 2026-09-26; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1657 per year. It increases by 20% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1965.4", "answer_3": "2000.4", "answer_2": "1325.6", "answer_1": "1988.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 258 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "120", "answer_2": "145", "answer_3": "137", "answer_0": "140"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $438. A fee of $73 is charged, then a deposit of $86 arrives. A second fee of $36 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "439", "answer_1": "428", "answer_2": "424", "answer_0": "415"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $23k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 8 grams of salt per serving and serves 7 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $535. A fee of $79 is charged, then a deposit of $66 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "467", "answer_3": "503", "answer_0": "483", "answer_1": "477"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 7 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 297 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "168", "answer_2": "180", "answer_3": "180", "answer_1": "164"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: move me to the pro plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1388 per year. It increases by 20% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1665.6", "answer_0": "1631.6", "answer_2": "1110.4", "answer_3": "1653.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = 2026-09-17; priority = low.\n\nClaim to verify: \"the owner is 2026-09-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.417", "p0": "0.583", "p3": "0.545", "p1": "0.417"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Thursday, December 24, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.625", "p0": "0.500", "p2": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Contract signed July 09, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 19", "date_3": "July 21", "date_2": "July 21", "date_0": "July 23"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "56", "answer_1": "20", "answer_3": "42", "answer_2": "14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $820. A fee of $35 is charged, then a deposit of $71 arrives. A second fee of $17 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "812", "answer_2": "813", "answer_1": "828", "answer_0": "839"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1969. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "56", "age_2": "58", "age_3": "55", "age_0": "57"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=737, unit_price=137, discount_pct=0, region=EU, sku_total=874 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "874", "v2": "10096", "v0": "100969", "v3": "100232"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $471 per year. It increases by 25% this year, then a $35 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "588.75", "answer_3": "572.75", "answer_0": "553.75", "answer_2": "353.25"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 2 shifts this week. A client ordered 241 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "46", "answer_1": "40", "answer_2": "42", "answer_0": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p1": "0.500", "p3": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.444", "p1": "0.500", "p2": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $38k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p0": "0.150", "p1": "0.200", "p3": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed December 15, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 18", "date_2": "December 16", "date_3": "December 20", "date_0": "December 18"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.222", "p1": "0.222", "p3": "0.750", "p0": "0.778"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 4 grams of salt per serving and serves 12 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = open; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $213 per year. It increases by 10% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "163.3", "answer_0": "198.3", "answer_2": "191.7", "answer_1": "234.3"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $609. A fee of $183 is charged, then a deposit of $44 arrives. A second fee of $91 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "377", "answer_2": "367", "answer_3": "383", "answer_0": "379"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: date = $130; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A shift runs from 07:00 to 13:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "5h45", "dur_1": "6h45", "dur_0": "6h00", "dur_2": "6h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $30k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed November 28, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "December 01", "date_0": "December 02", "date_2": "November 30", "date_3": "December 03"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = 2026-08-12; priority = low.\n\nClaim to verify: \"the date is 2026-08-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 16 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $584. A fee of $36 is charged, then a deposit of $75 arrives. A second fee of $18 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "628", "answer_0": "605", "answer_3": "601", "answer_1": "595"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=777, unit_price=690, discount_pct=0, region=EU, sku_total=1467 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1467", "v0": "536130", "v3": "536907", "v2": "53613"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I changed my mind, cancel it\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1211 per year. It increases by 5% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1271.55", "answer_0": "1225.55", "answer_2": "1150.45", "answer_3": "1245.55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = $314; priority = low.\n\nClaim to verify: \"the amount is $314\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = closed; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $976 per year. It increases by 15% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "829.6", "answer_3": "1102.4", "answer_0": "1086.4", "answer_1": "1122.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "51", "age_2": "53", "age_0": "52", "age_3": "50"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 216:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.467", "p1": "0.467", "p0": "0.533", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: amount = 2026-09-23; priority = high.\n\nClaim to verify: \"the amount is 2026-09-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $32k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_0": "30", "answer_3": "12", "answer_1": "22"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you thanks for the quick help\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $277. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = $474; priority = low.\n\nClaim to verify: \"the date is $474\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed April 24, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 02", "date_1": "April 28", "date_3": "April 30", "date_0": "April 30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed January 23, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "February 04", "date_2": "February 02", "date_3": "February 02", "date_1": "January 31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 604:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.300", "p2": "0.636", "p0": "0.364", "p1": "0.636"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 09:30 to 09:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h15", "dur_0": "-1h30", "dur_2": "0h00", "dur_1": "0h15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1981. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "43", "age_2": "46", "age_0": "45", "age_1": "44"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, September 21, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 362 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "192", "answer_2": "212", "answer_3": "184", "answer_1": "186"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed September 21, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 28", "date_1": "September 26", "date_3": "September 28", "date_2": "September 30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 5 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "25", "answer_2": "20", "answer_3": "20", "answer_0": "40"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 285 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "212", "answer_2": "241", "answer_1": "240", "answer_0": "224"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=945, unit_price=986, discount_pct=0, region=EU, sku_total=1931 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "930825", "v2": "93177", "v0": "931770", "v1": "1931"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.625", "p2": "0.375", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.462", "p2": "0.538", "p1": "0.538", "p3": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p2": "0.250", "p3": "0.125", "p1": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: date = open; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p3": "0.125", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.778", "p0": "0.222", "p3": "0.125", "p1": "0.778"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.167", "p0": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = $396; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed August 21, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 24", "date_0": "August 26", "date_2": "August 24", "date_3": "August 26"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1986. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "39", "age_3": "42", "age_2": "41", "age_0": "40"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 186 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "233", "answer_3": "219", "answer_0": "224", "answer_1": "222"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = $625; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed June 27, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 09", "date_1": "July 06", "date_2": "July 11", "date_3": "July 08"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed October 04, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 07", "date_0": "October 09", "date_1": "October 09", "date_3": "October 11"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_3": "42", "answer_1": "19", "answer_0": "54"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Contract signed August 13, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "August 19", "date_1": "August 17", "date_2": "August 17", "date_3": "August 19"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "72", "answer_2": "48", "answer_1": "51", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A subscription costs $959 per year. It increases by 25% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1198.75", "answer_0": "1181.75", "answer_3": "1174.75", "answer_2": "719.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.462", "p0": "0.538", "p3": "0.500", "p1": "0.462"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: date = 2026-12-24; priority = low.\n\nClaim to verify: \"the date is 2026-12-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = $251; priority = high.\n\nClaim to verify: \"the owner is $251\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 305 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "209", "answer_2": "229", "answer_0": "216", "answer_1": "228"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"track my shipment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1977. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "49", "age_3": "50", "age_1": "47", "age_0": "48"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 9 people. A shopper bought a 54 gram box and a 540 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 08:15 to 14:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "5h45", "dur_1": "6h15", "dur_0": "5h30", "dur_3": "5h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = $791; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1576 per year. It increases by 10% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1689.6", "answer_1": "1733.6", "answer_2": "1418.4", "answer_3": "1662.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = 2026-12-26; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p2": "0.417", "p0": "0.583", "p3": "0.545"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=244, unit_price=317, discount_pct=0, region=EU, sku_total=561 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "77348", "v2": "7734", "v3": "77104", "v1": "561"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1968. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "57", "age_2": "59", "age_3": "56", "age_0": "58"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Thursday, January 22, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.333", "p3": "0.167", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = closed; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1294 per year. It increases by 5% this year, then a $23 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1295.7", "answer_2": "1229.3", "answer_0": "1335.7", "answer_1": "1358.7"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = $846; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=559, unit_price=109, discount_pct=0, region=EU, sku_total=668 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "61490", "v1": "668", "v0": "60931", "v2": "6093"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1839 per year. It increases by 15% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2114.85", "answer_0": "2102.85", "answer_3": "2124.85", "answer_2": "1563.15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $335. A fee of $139 is charged, then a deposit of $43 arrives. A second fee of $69 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "158", "answer_2": "164", "answer_0": "170", "answer_1": "158"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, March 11, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1978. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "48", "age_2": "49", "age_1": "47", "age_3": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 3 grams of salt per serving and serves 5 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed May 10, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 17", "date_0": "May 19", "date_3": "May 19", "date_1": "May 17"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $224. A fee of $93 is charged, then a deposit of $66 arrives. A second fee of $46 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "122", "answer_1": "177", "answer_3": "134", "answer_0": "151"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.250", "p3": "0.083", "p1": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $318. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = $419; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.625", "p2": "0.375", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.167", "p1": "0.833", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 4 grams of salt per serving and serves 12 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 9 grams of salt per serving and serves 4 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 954:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 3 shifts this week. A client ordered 90 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "79", "answer_3": "88", "answer_1": "90", "answer_0": "75"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = 2026-12-11; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"forgot my password\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, July 07, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 6 hours per shift and runs 5 shifts this week. A client ordered 230 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_2": "113", "answer_3": "110", "answer_1": "137"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p1": "0.625", "p0": "0.500", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed December 02, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 07", "date_2": "December 05", "date_1": "December 05", "date_3": "December 07"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 6 grams of salt per serving and serves 11 people. A shopper bought a 66 gram box and a 660 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = 2026-03-12; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed July 08, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 19", "date_1": "July 17", "date_2": "July 23", "date_0": "July 21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed November 27, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 01", "date_0": "December 03", "date_3": "December 03", "date_1": "December 01"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $273. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1660 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1725", "answer_1": "1743", "answer_0": "1711", "answer_2": "1577"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An account starts at $491. A fee of $38 is charged, then a deposit of $63 arrives. A second fee of $19 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "497", "answer_1": "510", "answer_3": "495", "answer_2": "484"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=727, unit_price=965, discount_pct=0, region=EU, sku_total=1692 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "70155", "v3": "702282", "v1": "1692", "v0": "701555"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = $251; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: owner = $379; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=372, unit_price=508, discount_pct=0, region=EU, sku_total=880 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "18897", "v3": "188604", "v1": "880", "v0": "188976"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = $659; priority = low.\n\nClaim to verify: \"the status is $659\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Tuesday, May 12, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "24", "answer_1": "12", "answer_3": "16", "answer_2": "8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed December 10, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 22", "date_3": "December 22", "date_0": "December 24", "date_1": "December 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.375", "p2": "0.125", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1970. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "53", "age_1": "54", "age_2": "56", "age_0": "55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 14 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=340, unit_price=230, discount_pct=0, region=EU, sku_total=570 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "77860", "v0": "78200", "v2": "7820", "v1": "570"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Contract signed August 08, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 13", "date_3": "August 15", "date_0": "August 14", "date_2": "August 16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 393 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "192", "answer_1": "211", "answer_2": "179", "answer_3": "204"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p1": "0.625", "p2": "0.375", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.050", "p0": "0.100", "p1": "0.150"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = 2026-03-01; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: status = 2026-06-20; priority = high.\n\nClaim to verify: \"the status is 2026-06-20\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.417", "p2": "0.250", "p0": "0.333", "p3": "0.083"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.167", "p0": "0.250", "p1": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 694:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.467", "p1": "0.467", "p3": "0.500", "p0": "0.533"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $351. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $768. A fee of $174 is charged, then a deposit of $42 arrives. A second fee of $87 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "539", "answer_2": "520", "answer_0": "549", "answer_3": "568"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.125", "p1": "0.375", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = $688; priority = low.\n\nClaim to verify: \"the date is $688\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p3": "0.083", "p1": "0.250", "p0": "0.167"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1145 per year. It increases by 15% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1278.75", "answer_1": "1316.75", "answer_2": "973.25", "answer_0": "1271.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = 2026-02-06; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $246. A fee of $144 is charged, then a deposit of $90 arrives. A second fee of $72 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "96", "answer_1": "91", "answer_0": "120", "answer_2": "135"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 6 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 72 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "48", "answer_3": "51", "answer_2": "64", "answer_0": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed January 13, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 19", "date_3": "January 19", "date_2": "January 17", "date_1": "January 17"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: status = 2026-03-08; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.167", "p3": "0.083", "p1": "0.250", "p2": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: status = 2026-10-18; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "36", "answer_1": "30", "answer_0": "60", "answer_2": "24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=101, unit_price=475, discount_pct=0, region=EU, sku_total=576 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "4797", "v1": "576", "v3": "48076", "v0": "47975"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=270, unit_price=661, discount_pct=0, region=EU, sku_total=931 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "178200", "v1": "931", "v0": "178470", "v2": "17847"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p3": "0.125", "p0": "0.250", "p2": "0.125"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 5 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.444", "p0": "0.500", "p2": "0.500", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Form fields: date = 2026-05-02; priority = high.\n\nClaim to verify: \"the date is 2026-05-02\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1965. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "61", "age_0": "60", "age_1": "59", "age_3": "62"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: owner = 2026-08-03; priority = low.\n\nClaim to verify: \"the owner is 2026-08-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p3": "0.125", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=788, unit_price=747, discount_pct=0, region=EU, sku_total=1535 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1535", "v0": "588636", "v2": "58863", "v3": "589424"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 8 grams of salt per serving and serves 10 people. A shopper bought a 80 gram box and a 800 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.833", "p2": "0.500", "p0": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Monday, July 27, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 10:30 to 18:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "6h30", "dur_1": "7h30", "dur_0": "7h00", "dur_2": "7h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 07:45 to 08:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "0h45", "dur_1": "0h45", "dur_0": "0h15", "dur_3": "-1h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 4 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "27", "answer_3": "18", "answer_0": "42", "answer_2": "24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p0": "0.500", "p2": "0.333", "p1": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=224, unit_price=784, discount_pct=0, region=EU, sku_total=1008 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "175392", "v1": "1008", "v2": "17561", "v0": "175616"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 183 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "317", "answer_2": "343", "answer_3": "343", "answer_0": "336"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "20", "answer_2": "8", "answer_1": "13", "answer_0": "28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 9 grams of salt per serving and serves 8 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "36", "answer_1": "36", "answer_0": "66", "answer_2": "30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 517:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 97 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "126", "answer_2": "154", "answer_1": "161", "answer_0": "144"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 6 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.571", "p1": "0.571", "p3": "0.385", "p0": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "40", "answer_0": "72", "answer_1": "37", "answer_2": "32"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.500", "p3": "0.167", "p1": "0.833"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 383 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "54", "answer_1": "65", "answer_3": "60", "answer_2": "69"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 4 grams of salt per serving and serves 7 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $22k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Monday, July 06, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=302, unit_price=138, discount_pct=0, region=EU, sku_total=440 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "41978", "v1": "440", "v2": "4167", "v0": "41676"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = 2026-12-24; priority = high.\n\nClaim to verify: \"the owner is 2026-12-24\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, January 09, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = 2026-10-05; priority = high.\n\nClaim to verify: \"the status is 2026-10-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p0": "0.333", "p2": "0.667", "p1": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed November 13, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 18", "date_3": "November 20", "date_0": "November 20", "date_2": "November 18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $829 per year. It increases by 10% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "911.9", "answer_0": "878.9", "answer_3": "912.9", "answer_2": "746.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1576 per year. It increases by 25% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1182", "answer_1": "1970", "answer_3": "1933", "answer_0": "1924"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed April 22, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "May 03", "date_0": "May 05", "date_3": "May 03", "date_1": "May 01"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.625", "p3": "0.571", "p1": "0.375", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 4 shifts this week. A client ordered 133 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "155", "answer_3": "146", "answer_2": "135", "answer_0": "144"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = pending; priority = high.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = $507; priority = low.\n\nClaim to verify: \"the amount is $507\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.375", "p3": "0.125", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p3": "0.125", "p1": "0.500", "p0": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "65", "age_3": "66", "age_1": "63", "age_0": "64"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 18:30 to 20:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "0h45", "dur_2": "1h00", "dur_1": "1h30", "dur_3": "0h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_0": "28", "answer_1": "19", "answer_3": "12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 24 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p2": "0.600", "p3": "0.333", "p1": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=655, unit_price=713, discount_pct=0, region=EU, sku_total=1368 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "467015", "v2": "46701", "v3": "466360", "v1": "1368"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1931 per year. It increases by 10% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1737.9", "answer_3": "2094.1", "answer_0": "2110.1", "answer_1": "2124.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "41", "answer_0": "66", "answer_2": "36", "answer_3": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p3": "0.083", "p2": "0.167", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 3 grams of salt per serving and serves 12 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.571", "p1": "0.571", "p3": "0.333", "p0": "0.429"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "8", "answer_0": "12", "answer_2": "6", "answer_3": "6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1409 per year. It increases by 15% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1597.35", "answer_0": "1592.35", "answer_1": "1620.35", "answer_2": "1197.65"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.467", "p2": "0.500", "p0": "0.500", "p1": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: date = open; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 390:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "24", "answer_3": "36", "answer_0": "54", "answer_2": "18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p1": "0.625", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed July 04, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "July 10", "date_2": "July 12", "date_1": "July 09", "date_3": "July 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 11 people. A shopper bought a 99 gram box and a 990 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.333", "p0": "0.667", "p3": "0.600"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 408:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p2": "0.125", "p0": "0.250", "p3": "0.125"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=489, unit_price=928, discount_pct=0, region=EU, sku_total=1417 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1417", "v0": "453792", "v2": "45379", "v3": "453303"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1863 per year. It increases by 25% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2290.75", "answer_1": "2328.75", "answer_3": "2298.75", "answer_2": "1397.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 219 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "164", "answer_2": "150", "answer_3": "146", "answer_0": "160"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: date = $457; priority = high.\n\nClaim to verify: \"the date is $457\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: date = $113; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1980. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "43", "age_3": "42", "age_2": "45", "age_0": "44"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed September 26, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 30", "date_2": "October 03", "date_0": "October 01", "date_3": "October 02"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p3": "0.625", "p2": "0.333", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.500", "p2": "0.333", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $628. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=121, unit_price=503, discount_pct=0, region=EU, sku_total=624 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "60863", "v1": "624", "v2": "6086", "v3": "60742"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 189 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "80", "answer_2": "55", "answer_1": "54", "answer_0": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 14:45 to 15:30 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "0h30", "dur_2": "1h00", "dur_1": "0h45", "dur_3": "-1h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1890 per year. It increases by 20% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2246", "answer_1": "2268", "answer_2": "1512", "answer_3": "2253"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 11 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed October 19, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "October 21", "date_0": "October 23", "date_1": "October 23", "date_3": "October 25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = 2026-09-12; priority = high.\n\nClaim to verify: \"the date is 2026-09-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $37k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, August 14, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1285 per year. It increases by 25% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1545.25", "answer_2": "963.75", "answer_1": "1606.25", "answer_0": "1575.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1989. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "36", "age_2": "37", "age_3": "38", "age_1": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=802, unit_price=392, discount_pct=0, region=EU, sku_total=1194 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "31438", "v0": "314384", "v3": "315186", "v1": "1194"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 304 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "315", "answer_2": "318", "answer_3": "304", "answer_1": "303"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: date = 2026-10-13; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = $650; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 07:30 to 21:00 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "13h30", "dur_0": "12h45", "dur_3": "12h30", "dur_2": "13h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "42", "answer_1": "22", "answer_2": "18", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=380, unit_price=527, discount_pct=0, region=EU, sku_total=907 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "20026", "v1": "907", "v3": "199880", "v0": "200260"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = 2026-06-28; priority = low.\n\nClaim to verify: \"the status is 2026-06-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1764 per year. It increases by 15% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2034.6", "answer_2": "1499.4", "answer_0": "2006.6", "answer_1": "2028.6"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Please the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.200", "p3": "0.778", "p1": "0.200", "p0": "0.800"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $567. A fee of $101 is charged, then a deposit of $70 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "486", "answer_3": "475", "answer_2": "461", "answer_1": "493"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = open; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $396. A fee of $97 is charged, then a deposit of $76 arrives. A second fee of $48 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "306", "answer_1": "334", "answer_2": "301", "answer_0": "327"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 9 grams of salt per serving and serves 10 people. A shopper bought a 90 gram box and a 900 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $820. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = open; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.429", "p0": "0.571", "p2": "0.429", "p3": "0.538"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: owner = 2026-12-23; priority = low.\n\nClaim to verify: \"the owner is 2026-12-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed March 28, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "April 09", "date_3": "April 06", "date_1": "April 04", "date_0": "April 07"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, June 12, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=574, unit_price=228, discount_pct=0, region=EU, sku_total=802 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "13087", "v0": "130872", "v3": "131446", "v1": "802"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1176 per year. It increases by 10% this year, then a $41 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1058.4", "answer_0": "1252.6", "answer_1": "1293.6", "answer_3": "1290.6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Thursday, June 25, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1965. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "60", "age_3": "58", "age_2": "61", "age_1": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: date = $59; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 5 grams of salt per serving and serves 5 people. A shopper bought a 25 gram box and a 250 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 964:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 421:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = 2026-04-05; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = $722; priority = low.\n\nClaim to verify: \"the date is $722\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.250", "p1": "0.417", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 12 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1968. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "59", "age_0": "57", "age_2": "58", "age_1": "56"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $879. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"my package was damaged\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 5 days. Team B finishes 4 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "20", "answer_0": "30", "answer_1": "14", "answer_2": "10"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed September 07, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "September 16", "date_3": "September 14", "date_0": "September 14", "date_1": "September 12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_0": "77", "answer_1": "47", "answer_2": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed June 01, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 06", "date_0": "June 08", "date_1": "June 06", "date_3": "June 08"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = 2026-06-05; priority = high.\n\nClaim to verify: \"the amount is 2026-06-05\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=699, unit_price=349, discount_pct=0, region=EU, sku_total=1048 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "244650", "v0": "243951", "v2": "24395", "v1": "1048"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.050", "p1": "0.150", "p0": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 22 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.583", "p3": "0.364", "p2": "0.583", "p0": "0.417"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=363, unit_price=530, discount_pct=0, region=EU, sku_total=893 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "19239", "v1": "893", "v0": "192390", "v3": "192753"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.167", "p1": "0.250", "p2": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.500", "p2": "0.167", "p0": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed February 06, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 17", "date_0": "February 19", "date_2": "February 21", "date_1": "February 15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "32", "answer_3": "48", "answer_0": "80", "answer_1": "38"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1990. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "34", "age_3": "36", "age_2": "35", "age_1": "33"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 947:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = $807; priority = low.\n\nClaim to verify: \"the status is $807\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "28", "answer_3": "14", "answer_2": "14", "answer_1": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.250", "p1": "0.333", "p3": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.667", "p2": "0.333", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=270, unit_price=914, discount_pct=0, region=EU, sku_total=1184 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1184", "v0": "246780", "v2": "24678", "v3": "246510"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.615", "p2": "0.615", "p0": "0.385", "p3": "0.333"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = 2026-06-19; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $680. A fee of $190 is charged, then a deposit of $24 arrives. A second fee of $95 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "442", "answer_0": "419", "answer_3": "438", "answer_1": "410"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 422:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1965. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "60", "age_2": "62", "age_0": "61", "age_3": "59"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $737. A fee of $100 is charged, then a deposit of $77 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "686", "answer_3": "690", "answer_0": "664", "answer_2": "686"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 9 grams of salt per serving and serves 4 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $510 per year. It increases by 25% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "382.5", "answer_3": "644.5", "answer_1": "637.5", "answer_0": "608.5"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed October 25, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 08", "date_0": "November 06", "date_3": "November 06", "date_1": "November 04"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.615", "p0": "0.385", "p1": "0.615", "p3": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1975. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "53", "age_1": "50", "age_0": "51", "age_2": "52"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "55", "answer_3": "25", "answer_2": "30", "answer_1": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1332 per year. It increases by 5% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1265.4", "answer_0": "1351.6", "answer_1": "1398.6", "answer_3": "1312.6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 6 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1110 per year. It increases by 25% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1387.5", "answer_2": "832.5", "answer_3": "1332.5", "answer_0": "1358.5"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.300", "p3": "0.667", "p0": "0.700", "p2": "0.300"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $925 per year. It increases by 25% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1144.25", "answer_2": "693.75", "answer_3": "1158.25", "answer_1": "1156.25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=104, unit_price=394, discount_pct=0, region=EU, sku_total=498 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "4097", "v3": "41080", "v1": "498", "v0": "40976"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $544 per year. It increases by 25% this year, then a $17 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "663", "answer_3": "694", "answer_2": "408", "answer_1": "680"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $222. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 171 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "59", "answer_1": "83", "answer_2": "84", "answer_0": "70"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 7 grams of salt per serving and serves 10 people. A shopper bought a 70 gram box and a 700 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $211. A fee of $111 is charged, then a deposit of $49 arrives. A second fee of $55 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "111", "answer_2": "107", "answer_0": "94", "answer_3": "107"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: date = ops; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: date = $195; priority = low.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1811 per year. It increases by 5% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1901.55", "answer_2": "1720.45", "answer_0": "1858.55", "answer_3": "1896.55"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = $674; priority = low.\n\nClaim to verify: \"the date is $674\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 356:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_0": "60", "answer_1": "32", "answer_2": "25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1988. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "36", "age_2": "38", "age_3": "39", "age_0": "37"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 10 people. A shopper bought a 50 gram box and a 500 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1995 per year. It increases by 10% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "2151.5", "answer_1": "2194.5", "answer_2": "1795.5", "answer_0": "2184.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 366:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed May 06, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 18", "date_2": "May 20", "date_1": "May 14", "date_3": "May 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 9 grams of salt per serving and serves 9 people. A shopper bought a 81 gram box and a 810 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p2": "0.571", "p0": "0.429", "p1": "0.571"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=427, unit_price=334, discount_pct=0, region=EU, sku_total=761 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "142618", "v1": "761", "v2": "14261", "v3": "142191"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.417", "p0": "0.333", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "51", "age_0": "52", "age_2": "53", "age_3": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, April 12, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=722, unit_price=245, discount_pct=0, region=EU, sku_total=967 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "176890", "v1": "967", "v3": "176168", "v2": "17689"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1962. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "62", "age_1": "61", "age_2": "63", "age_3": "60"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 3 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=700, unit_price=589, discount_pct=0, region=EU, sku_total=1289 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "41230", "v3": "411600", "v1": "1289", "v0": "412300"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = $484; priority = high.\n\nClaim to verify: \"the status is $484\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1975. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "52", "age_3": "53", "age_1": "50", "age_0": "51"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $897. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $953 per year. It increases by 15% this year, then a $30 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1095.95", "answer_0": "1065.95", "answer_3": "1078.95", "answer_2": "810.05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 21 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 10:15 to 15:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "5h00", "dur_3": "4h15", "dur_1": "5h15", "dur_0": "4h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 9 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Thursday, June 11, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = 2026-02-24; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = $658; priority = high.\n\nClaim to verify: \"the date is $658\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $463. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 319 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "354", "answer_1": "333", "answer_3": "352", "answer_0": "336"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = closed; priority = high.\n\nClaim to verify: \"the owner is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 7 days. Team B finishes 2 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "49", "answer_3": "14", "answer_1": "37", "answer_2": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 3 days. Team B finishes 5 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "14", "answer_3": "15", "answer_2": "9", "answer_0": "24"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=741, unit_price=658, discount_pct=0, region=EU, sku_total=1399 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "488319", "v0": "487578", "v1": "1399", "v2": "48757"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.667", "p2": "0.333", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = closed; priority = low.\n\nClaim to verify: \"the date is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: owner = 2026-10-18; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = 2026-10-25; priority = low.\n\nClaim to verify: \"the status is 2026-10-25\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $299. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1983. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "41", "age_3": "44", "age_0": "42", "age_2": "43"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p3": "0.050", "p1": "0.250", "p0": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Monday, October 19, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: date = $250; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $171. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1974. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "52", "age_1": "50", "age_0": "51", "age_3": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1402 per year. It increases by 5% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1424.1", "answer_2": "1331.9", "answer_1": "1472.1", "answer_3": "1437.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 232 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "171", "answer_0": "189", "answer_3": "175", "answer_2": "199"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1973. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "53", "age_2": "54", "age_3": "55", "age_1": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed November 09, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "November 19", "date_2": "November 21", "date_0": "November 23", "date_3": "November 21"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Sunday, December 27, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p2": "0.375", "p0": "0.500", "p1": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 15 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = 2026-02-14; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $36k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 2 grams of salt per serving and serves 12 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p1": "0.500", "p2": "0.250", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $555 per year. It increases by 15% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "638.25", "answer_2": "471.75", "answer_0": "591.25", "answer_3": "572.25"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = pending; priority = low.\n\nClaim to verify: \"the date is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.417", "p3": "0.083", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p0": "0.200", "p3": "0.050", "p1": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 739:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.333", "p0": "0.500", "p1": "0.667", "p3": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.364", "p1": "0.364", "p3": "0.600", "p0": "0.636"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $626. A fee of $50 is charged, then a deposit of $33 arrives. A second fee of $25 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "584", "answer_3": "569", "answer_1": "564", "answer_2": "568"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 386 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "99", "answer_1": "111", "answer_3": "113", "answer_0": "96"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "63", "age_2": "65", "age_0": "64", "age_3": "66"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 4 grams of salt per serving and serves 4 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"text mom I am running late\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = 2026-04-08; priority = high.\n\nClaim to verify: \"the status is 2026-04-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p3": "0.167", "p0": "0.500", "p2": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A subscription costs $571 per year. It increases by 25% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "428.25", "answer_0": "703.75", "answer_1": "713.75", "answer_3": "729.75"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Friday, August 28, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 8 crates per hour. It operates 5 hours per shift and runs 2 shifts this week. A client ordered 244 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "65", "answer_1": "67", "answer_0": "80", "answer_3": "74"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = team-b; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 451:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $732. A fee of $58 is charged, then a deposit of $69 arrives. A second fee of $29 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "714", "answer_3": "739", "answer_1": "690", "answer_2": "719"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "35", "answer_2": "14", "answer_1": "19", "answer_0": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $973 per year. It increases by 5% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "924.35", "answer_3": "1003.65", "answer_1": "1021.65", "answer_0": "978.65"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Sunday, July 12, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please the product is defective\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = 2026-09-24; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"send it to my new address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 324 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "104", "answer_1": "73", "answer_2": "68", "answer_0": "84"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "16", "answer_1": "42", "answer_2": "40", "answer_0": "56"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = closed; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Wednesday, June 03, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p2": "0.167", "p1": "0.333", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 09:45 to 21:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "11h15", "dur_1": "11h30", "dur_0": "10h45", "dur_3": "10h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed February 24, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 04", "date_3": "March 04", "date_1": "March 02", "date_2": "March 06"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1972. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "51", "age_1": "52", "age_2": "54", "age_0": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 8 grams of salt per serving and serves 7 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: date = open; priority = high.\n\nClaim to verify: \"the date is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.667", "p3": "0.200", "p2": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 233:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $258. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $669 per year. It increases by 10% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "706.9", "answer_3": "719.9", "answer_1": "735.9", "answer_2": "602.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, where is my package\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 2 shifts this week. A client ordered 360 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "36", "answer_1": "35", "answer_2": "33", "answer_0": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = 2026-05-12; priority = low.\n\nClaim to verify: \"the owner is 2026-05-12\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 7 grams of salt per serving and serves 6 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p3": "0.625", "p1": "0.333", "p2": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed January 23, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 28", "date_2": "January 28", "date_3": "January 30", "date_0": "January 30"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p0": "0.100", "p2": "0.050", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Sunday, July 19, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 118 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "40", "answer_2": "50", "answer_0": "30", "answer_3": "46"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p3": "0.500", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = $725; priority = high.\n\nClaim to verify: \"the owner is $725\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, January 27, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Wednesday, December 16, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Sunday, June 07, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: date = 2026-07-22; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1987. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "36", "age_3": "35", "age_2": "38", "age_0": "37"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1973. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "49", "age_1": "50", "age_0": "51", "age_2": "52"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $654 per year. It increases by 20% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "523.2", "answer_0": "774.8", "answer_3": "799.8", "answer_1": "784.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1960. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "64", "age_3": "67", "age_0": "65", "age_2": "66"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: status = pending; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 61 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "275", "answer_3": "291", "answer_2": "287", "answer_0": "280"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = 2026-11-26; priority = high.\n\nClaim to verify: \"the amount is 2026-11-26\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p2": "0.667", "p0": "0.333", "p3": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 923:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 5 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed August 16, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 22", "date_3": "August 24", "date_2": "August 26", "date_0": "August 24"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = 2026-08-13; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to unsubscribe\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A subscription costs $1085 per year. It increases by 20% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1317", "answer_2": "868", "answer_0": "1281", "answer_1": "1302"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 8 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 754:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = 2026-12-23; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "24", "answer_0": "36", "answer_2": "12", "answer_1": "16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p2": "0.250", "p1": "0.417", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p3": "0.083", "p0": "0.250", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.375", "p1": "0.500", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.333", "p3": "0.083", "p1": "0.417"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "23", "answer_2": "16", "answer_3": "28", "answer_0": "44"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $809. A fee of $167 is charged, then a deposit of $31 arrives. A second fee of $83 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "582", "answer_0": "590", "answer_2": "593", "answer_1": "573"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.250", "p1": "0.333", "p3": "0.083"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p1": "0.250", "p2": "0.083", "p0": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed February 21, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "February 26", "date_0": "February 25", "date_1": "February 24", "date_2": "February 23"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "36", "answer_3": "30", "answer_2": "30", "answer_0": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: date = $85; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.545", "p2": "0.455", "p3": "0.500", "p1": "0.455"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 8 grams of salt per serving and serves 3 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = $830; priority = high.\n\nClaim to verify: \"the status is $830\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_3": "12", "answer_0": "36", "answer_1": "27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.125", "p1": "0.375", "p0": "0.250", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed October 26, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 07", "date_1": "November 05", "date_0": "November 09", "date_2": "November 11"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $433. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: give me a refund\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $37k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1959. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "65", "age_0": "66", "age_3": "68", "age_2": "67"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $223. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 9 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 205:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = ops; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 7 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "14", "answer_0": "63", "answer_1": "21", "answer_3": "49"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p0": "0.333", "p2": "0.667", "p1": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $154. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $767 per year. It increases by 20% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "896.4", "answer_1": "920.4", "answer_0": "875.4", "answer_2": "613.6"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 8 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 204 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "96", "answer_2": "92", "answer_1": "82", "answer_3": "100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.250", "p3": "0.125", "p0": "0.375"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Contract signed September 22, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "October 01", "date_1": "September 29", "date_2": "September 29", "date_0": "October 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = pending; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.150", "p1": "0.200", "p2": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = 2026-06-13; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Thursday, November 12, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 17:30 to 21:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "3h45", "dur_2": "3h45", "dur_0": "3h30", "dur_3": "2h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 5 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 287 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_2": "46", "answer_3": "48", "answer_1": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 4 shifts this week. A client ordered 217 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "78", "answer_0": "96", "answer_3": "91", "answer_1": "81"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p0": "0.150", "p2": "0.100", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = $400; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.750", "p3": "0.143", "p0": "0.250", "p2": "0.750"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, May 25, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=454, unit_price=187, discount_pct=0, region=EU, sku_total=641 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "8489", "v0": "84898", "v3": "85352", "v1": "641"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"remind me at 8am\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 800:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $323. A fee of $157 is charged, then a deposit of $23 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "86", "answer_0": "111", "answer_1": "107", "answer_2": "83"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = 2026-11-04; priority = low.\n\nClaim to verify: \"the status is 2026-11-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = 2026-07-08; priority = low.\n\nClaim to verify: \"the status is 2026-07-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed March 25, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "April 07", "date_2": "April 05", "date_1": "April 03", "date_3": "April 05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed February 20, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 01", "date_3": "March 03", "date_0": "March 05", "date_2": "March 07"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed July 13, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "July 21", "date_1": "July 21", "date_3": "July 23", "date_0": "July 23"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1961. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "65", "age_0": "64", "age_3": "66", "age_1": "63"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1959. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "68", "age_1": "66", "age_0": "67", "age_3": "65"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 3 grams of salt per serving and serves 3 people. A shopper bought a 9 gram box and a 90 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $26k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1406 per year. It increases by 5% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1462.3", "answer_3": "1455.3", "answer_1": "1476.3", "answer_2": "1335.7"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $10k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 2 tasks per day for 3 days. Team B finishes 2 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "12", "answer_3": "6", "answer_1": "8", "answer_2": "6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: amount = $595; priority = low.\n\nClaim to verify: \"the amount is $595\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = 2026-03-20; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Wednesday, November 25, 2026. The SLA requires a response within 9 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p0": "0.333", "p3": "0.083", "p1": "0.417"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 161 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "106", "answer_0": "96", "answer_3": "82", "answer_2": "87"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.050", "p0": "0.100", "p1": "0.150", "p3": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 868:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1989. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "34", "age_0": "36", "age_2": "37", "age_1": "35"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1980 per year. It increases by 25% this year, then a $33 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1485", "answer_1": "2475", "answer_3": "2480", "answer_0": "2442"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: amount = $819; priority = low.\n\nClaim to verify: \"the amount is $819\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.615", "p3": "0.583", "p1": "0.385", "p2": "0.385"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1960. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "66", "age_3": "63", "age_1": "64", "age_0": "65"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1898 per year. It increases by 20% this year, then a $45 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "2277.6", "answer_2": "1518.4", "answer_3": "2247.6", "answer_0": "2232.6"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p2": "0.050", "p1": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 6 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.333", "p2": "0.333", "p0": "0.667", "p3": "0.625"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A shift runs from 16:15 to 20:45 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "3h30", "dur_0": "4h15", "dur_2": "4h30", "dur_1": "4h30"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1295 per year. It increases by 25% this year, then a $21 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1618.75", "answer_3": "1573.75", "answer_2": "971.25", "answer_0": "1597.75"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed December 25, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "January 01", "date_2": "January 03", "date_1": "December 30", "date_3": "January 01"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = closed; priority = low.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.375", "p2": "0.250", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1967. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "61", "age_2": "60", "age_1": "58", "age_0": "59"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1852 per year. It increases by 15% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1574.2", "answer_3": "2081.8", "answer_1": "2129.8", "answer_0": "2104.8"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 734:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = closed; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A subscription costs $1842 per year. It increases by 20% this year, then a $43 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2167.4", "answer_1": "2210.4", "answer_3": "2174.4", "answer_2": "1473.6"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: owner = 2026-11-22; priority = low.\n\nClaim to verify: \"the owner is 2026-11-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p0": "0.333", "p1": "0.667", "p3": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1406 per year. It increases by 5% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1335.7", "answer_3": "1455.3", "answer_1": "1476.3", "answer_0": "1445.3"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = 2026-11-13; priority = low.\n\nClaim to verify: \"the status is 2026-11-13\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: owner = 2026-11-22; priority = low.\n\nClaim to verify: \"the owner is 2026-11-22\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1975. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "49", "age_0": "50", "age_2": "51", "age_3": "48"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.150", "p0": "0.100", "p2": "0.050", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = 2026-07-03; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "54", "age_2": "53", "age_0": "52", "age_1": "51"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 534:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: owner = open; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Customer data corrupted in production. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p3": "0.429", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Form fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $348. A fee of $174 is charged, then a deposit of $35 arrives. A second fee of $87 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "114", "answer_1": "96", "answer_3": "119", "answer_0": "122"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = 2026-01-12; priority = high.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1976. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "50", "age_3": "52", "age_2": "51", "age_1": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: date = 2026-11-17; priority = low.\n\nClaim to verify: \"the date is 2026-11-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1980. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "45", "age_0": "44", "age_1": "43", "age_3": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.250", "p3": "0.083", "p1": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $368. A fee of $156 is charged, then a deposit of $38 arrives. A second fee of $78 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "151", "answer_2": "162", "answer_3": "152", "answer_0": "172"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: status = ops; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $209. A fee of $129 is charged, then a deposit of $44 arrives. A second fee of $64 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_3": "71", "answer_2": "87", "answer_1": "77"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 3 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "14", "answer_1": "17", "answer_3": "21", "answer_0": "35"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 5 tasks per day for 5 days. Team B finishes 7 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "32", "answer_2": "25", "answer_0": "60", "answer_3": "35"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 8 grams of salt per serving and serves 3 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, August 11, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $1145 per year. It increases by 20% this year, then a $34 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1374", "answer_2": "916", "answer_3": "1350", "answer_0": "1340"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p2": "0.077", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed August 18, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "August 28", "date_1": "August 26", "date_2": "August 26", "date_0": "August 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed November 06, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 18", "date_1": "November 14", "date_2": "November 20", "date_3": "November 16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A shift runs from 16:15 to 21:15 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "5h00", "dur_0": "4h45", "dur_1": "5h00", "dur_3": "4h00"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p1": "0.500", "p0": "0.500", "p3": "0.462"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = open; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "cancel_order": "Route to the 'cancel order' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 987:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.375", "p3": "0.125", "p2": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p3": "0.050", "p0": "0.200", "p1": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 2 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p0": "0.400", "p2": "0.600", "p3": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1958. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "65", "age_2": "67", "age_3": "68", "age_0": "66"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 4 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.556", "p2": "0.556", "p0": "0.444", "p3": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 5 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 147 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "160", "answer_0": "175", "answer_2": "178", "answer_3": "188"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p2": "0.333", "p0": "0.500", "p1": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.375", "p3": "0.125", "p2": "0.125", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Contract signed June 16, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "June 27", "date_1": "June 23", "date_3": "June 25", "date_0": "June 25"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 294 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "31", "answer_2": "28", "answer_1": "8", "answer_0": "20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1973. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "50", "age_2": "53", "age_1": "51", "age_0": "52"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1646 per year. It increases by 15% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1853.9", "answer_3": "1874.9", "answer_2": "1399.1", "answer_1": "1892.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 351 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "23", "answer_3": "16", "answer_0": "32", "answer_2": "40"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1978. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "46", "age_1": "45", "age_2": "47", "age_3": "44"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = 2026-09-11; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = 2026-05-06; priority = high.\n\nClaim to verify: \"the status is 2026-05-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "18", "answer_2": "12", "answer_3": "24", "answer_0": "36"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p2": "0.615", "p0": "0.385", "p1": "0.615"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p1": "0.200", "p3": "0.050", "p2": "0.100"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = $490; priority = high.\n\nClaim to verify: \"the status is $490\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=816, unit_price=260, discount_pct=0, region=EU, sku_total=1076 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "21216", "v1": "1076", "v0": "212160", "v3": "212976"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = $606; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=281, unit_price=827, discount_pct=0, region=EU, sku_total=1108 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1108", "v0": "232387", "v2": "23238", "v3": "232106"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 15:00 to 20:15 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "4h30", "dur_1": "5h15", "dur_3": "4h15", "dur_2": "5h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed October 27, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "November 02", "date_1": "October 31", "date_3": "November 02", "date_2": "October 31"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1987. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "37", "age_3": "40", "age_0": "38", "age_2": "39"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Thursday, January 15, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.125", "p0": "0.375", "p1": "0.500", "p2": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=604, unit_price=449, discount_pct=0, region=EU, sku_total=1053 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "27119", "v3": "270592", "v0": "271196", "v1": "1053"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 6 shifts this week. A client ordered 341 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "207", "answer_0": "216", "answer_2": "227", "answer_1": "219"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p3": "0.050", "p0": "0.200", "p2": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1622 per year. It increases by 10% this year, then a $46 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1784.2", "answer_2": "1459.8", "answer_3": "1747.2", "answer_0": "1738.2"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Wednesday, May 13, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: amount = $651; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1978. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "46", "age_1": "45", "age_3": "44", "age_2": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = team-a; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 521:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Friday, May 01, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p2": "0.250", "p1": "0.500", "p3": "0.125"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 9 crates per hour. It operates 7 hours per shift and runs 3 shifts this week. A client ordered 281 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "178", "answer_0": "189", "answer_1": "209", "answer_2": "197"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $632. A fee of $35 is charged, then a deposit of $50 arrives. A second fee of $17 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "628", "answer_0": "630", "answer_3": "607", "answer_2": "650"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: PDF export missing images; copy-paste works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = open; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.286", "p3": "0.667", "p2": "0.286", "p0": "0.714"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1972. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "50", "age_0": "52", "age_1": "51", "age_2": "53"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.167", "p1": "0.833", "p0": "0.667", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 8 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 600:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, December 14, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 4 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 152 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "192", "answer_3": "178", "answer_1": "210", "answer_2": "173"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1112 per year. It increases by 25% this year, then a $22 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "834", "answer_1": "1390", "answer_3": "1381", "answer_0": "1368"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"tell Sarah the meeting moved\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = 2026-01-28; priority = low.\n\nClaim to verify: \"the amount is 2026-01-28\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, I need a ticket to Chicago next Friday\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "refund": "Route to the 'refund' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $676. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $447. A fee of $192 is charged, then a deposit of $32 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "191", "answer_1": "215", "answer_2": "209", "answer_3": "202"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 25 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $395. A fee of $185 is charged, then a deposit of $72 arrives. A second fee of $92 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "165", "answer_1": "201", "answer_0": "190", "answer_2": "187"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 13 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Saturday, March 07, 2026. The SLA requires a response within 16 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $19k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = team-b; priority = high.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 8 grams of salt per serving and serves 7 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.250", "p1": "0.417", "p0": "0.333", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please question about my invoice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $564. A fee of $105 is charged, then a deposit of $30 arrives. A second fee of $52 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "437", "answer_1": "448", "answer_2": "450", "answer_3": "418"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = 2026-06-03; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, February 10, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1971. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "56", "age_3": "57", "age_1": "54", "age_0": "55"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1045 per year. It increases by 15% this year, then a $28 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1164.75", "answer_1": "1201.75", "answer_2": "888.25", "answer_0": "1173.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.019", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 923:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.700", "p2": "0.300", "p3": "0.667", "p1": "0.300"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $40k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed October 19, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "October 26", "date_3": "October 26", "date_1": "October 24", "date_2": "October 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Saturday, November 28, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $729 per year. It increases by 20% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "874.8", "answer_2": "583.2", "answer_0": "824.8", "answer_3": "788.8"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: date = closed; priority = high.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=924, unit_price=116, discount_pct=0, region=EU, sku_total=1040 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "10718", "v3": "108108", "v0": "107184", "v1": "1040"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A warehouse packs 4 crates per hour. It operates 4 hours per shift and runs 3 shifts this week. A client ordered 158 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "34", "answer_3": "52", "answer_0": "48", "answer_2": "31"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 3 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.600", "p3": "0.500", "p2": "0.400", "p1": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to file a complaint\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Please this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $1530 per year. It increases by 15% this year, then a $38 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1757.5", "answer_0": "1721.5", "answer_2": "1300.5", "answer_1": "1759.5"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $134. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = 2026-10-23; priority = high.\n\nClaim to verify: \"the amount is 2026-10-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "8", "answer_3": "8", "answer_1": "10", "answer_0": "16"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 4 grams of salt per serving and serves 5 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: date = 2026-07-27; priority = high.\n\nClaim to verify: \"the date is 2026-07-27\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $4k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $566. A fee of $47 is charged, then a deposit of $23 arrives. A second fee of $23 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "504", "answer_0": "519", "answer_3": "502", "answer_2": "491"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 6 tasks per day for 7 days. Team B finishes 5 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "77", "answer_2": "42", "answer_3": "35", "answer_1": "47"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 4 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 172 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "29", "answer_0": "48", "answer_3": "31", "answer_1": "32"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $444. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.286", "p3": "0.667", "p0": "0.714", "p1": "0.286"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.273", "p0": "0.727", "p1": "0.273", "p3": "0.700"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=432, unit_price=750, discount_pct=0, region=EU, sku_total=1182 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "32400", "v1": "1182", "v0": "324000", "v3": "324432"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, May 26, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 3 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.500", "p1": "0.500", "p3": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = team-b; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 5 people. A shopper bought a 35 gram box and a 350 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=566, unit_price=703, discount_pct=0, region=EU, sku_total=1269 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "397898", "v1": "1269", "v3": "398464", "v2": "39789"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.375", "p0": "0.500", "p3": "0.125", "p1": "0.625"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incident report: Button misaligned on mobile. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p0": "0.200", "p1": "0.250", "p2": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=221, unit_price=214, discount_pct=0, region=EU, sku_total=435 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "47073", "v2": "4729", "v0": "47294", "v1": "435"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $875. A fee of $97 is charged, then a deposit of $89 arrives. A second fee of $48 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "819", "answer_1": "789", "answer_2": "799", "answer_3": "840"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1977. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "51", "age_1": "48", "age_2": "50", "age_0": "49"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 525:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reserve a seat on the morning flight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1963. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "64", "age_3": "65", "age_0": "63", "age_1": "62"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p1": "0.200", "p0": "0.150", "p2": "0.100"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want my money back\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.250", "p1": "0.667", "p2": "0.667"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_2": "12", "answer_1": "16", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 7 grams of salt per serving and serves 8 people. A shopper bought a 56 gram box and a 560 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"why was I charged twice\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "technical_issue": "Route to the 'technical issue' handler", "send_message": "Route to the 'send message' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I cant sign in, need a password reset\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "weather": "Route to the 'weather' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 8 grams of salt per serving and serves 11 people. A shopper bought a 88 gram box and a 880 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 8 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p0": "0.800", "p3": "0.778", "p2": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=202, unit_price=907, discount_pct=0, region=EU, sku_total=1109 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "183214", "v3": "183012", "v1": "1109", "v2": "18321"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A shift runs from 06:15 to 16:00 with a 15-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "9h45", "dur_3": "8h45", "dur_1": "9h45", "dur_0": "9h30"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=525, unit_price=295, discount_pct=0, region=EU, sku_total=820 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "155400", "v2": "15487", "v0": "154875", "v1": "820"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you switch to the business plan\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = team-a; priority = high.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 485:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"feature is broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Saturday, June 13, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 9 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 257 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "344", "answer_1": "316", "answer_3": "335", "answer_0": "324"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = 2026-04-23; priority = high.\n\nClaim to verify: \"the amount is 2026-04-23\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - what is the weather today\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you the item arrived broken\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "book_flight": "Route to the 'book flight' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, October 15, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 950:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $678. A fee of $140 is charged, then a deposit of $24 arrives. A second fee of $70 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "488", "answer_0": "492", "answer_3": "469", "answer_2": "497"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "13", "answer_3": "20", "answer_2": "8", "answer_0": "28"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "66", "answer_1": "36", "answer_3": "36", "answer_2": "30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = $318; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 4 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "12", "answer_1": "16", "answer_0": "28", "answer_3": "16"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1292 per year. It increases by 15% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "1485.8", "answer_2": "1098.2", "answer_0": "1453.8", "answer_3": "1491.8"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 218 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "193", "answer_3": "166", "answer_0": "180", "answer_1": "165"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 818:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 617:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "28", "answer_2": "24", "answer_3": "32", "answer_0": "56"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - refund me for the broken item\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 6 hours per shift and runs 6 shifts this week. A client ordered 273 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "198", "answer_3": "186", "answer_2": "187", "answer_0": "180"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "36", "answer_1": "18", "answer_0": "48", "answer_2": "12"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1985. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "40", "age_0": "41", "age_2": "42", "age_3": "43"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=673, unit_price=679, discount_pct=0, region=EU, sku_total=1352 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1352", "v3": "456294", "v2": "45696", "v0": "456967"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, August 23, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $30k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 8 days. Team B finishes 2 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "16", "answer_0": "40", "answer_1": "26", "answer_2": "24"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 5 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.455", "p2": "0.545", "p1": "0.545", "p3": "0.400"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"send a message to John\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A recipe needs 4 grams of salt per serving and serves 8 people. A shopper bought a 32 gram box and a 320 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 12 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $743. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = 2026-08-20; priority = low.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 143 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "66", "answer_3": "35", "answer_0": "54", "answer_2": "41"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1966. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_3": "56", "age_2": "59", "age_1": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 5 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_3": "20", "answer_0": "36", "answer_1": "21"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 27 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.100", "p0": "0.150", "p1": "0.200", "p3": "0.050"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $463. A fee of $114 is charged, then a deposit of $50 arrives. A second fee of $57 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "319", "answer_2": "319", "answer_3": "334", "answer_0": "342"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=818, unit_price=696, discount_pct=0, region=EU, sku_total=1514 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "570146", "v2": "56932", "v0": "569328", "v1": "1514"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 7 grams of salt per serving and serves 4 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 9 grams of salt per serving and serves 8 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: amount = $225; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1989. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "37", "age_2": "36", "age_0": "35", "age_1": "34"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 5 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "45", "answer_3": "25", "answer_2": "20", "answer_1": "25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: status = 2026-06-22; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.150", "p2": "0.100", "p3": "0.050", "p1": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 5 people. A shopper bought a 30 gram box and a 300 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Record fields: date = team-a; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1988. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "38", "age_1": "35", "age_2": "37", "age_0": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 5 crates per hour. It operates 8 hours per shift and runs 3 shifts this week. A client ordered 131 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_3": "139", "answer_2": "129", "answer_1": "132"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed December 04, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 18", "date_3": "December 16", "date_2": "December 20", "date_1": "December 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.222", "p1": "0.222", "p3": "0.750", "p0": "0.778"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $356. A fee of $150 is charged, then a deposit of $39 arrives. A second fee of $75 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "162", "answer_3": "197", "answer_2": "145", "answer_0": "170"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: owner = team-a; priority = low.\n\nClaim to verify: \"the owner is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1474 per year. It increases by 5% this year, then a $37 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1400.3", "answer_3": "1499.7", "answer_0": "1510.7", "answer_1": "1547.7"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 488:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Payment retries fail for a subset of users. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: amount = $702; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $676. A fee of $68 is charged, then a deposit of $38 arrives. A second fee of $34 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "620", "answer_1": "623", "answer_0": "612", "answer_2": "620"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A shift runs from 07:00 to 18:45 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "11h00", "dur_1": "11h45", "dur_2": "11h30", "dur_3": "10h45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: status = open; priority = high.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Form fields: owner = 2026-12-11; priority = high.\n\nClaim to verify: \"the owner is 2026-12-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $408. A fee of $104 is charged, then a deposit of $68 arrives. A second fee of $52 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "303", "answer_2": "317", "answer_0": "320", "answer_1": "341"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=848, unit_price=144, discount_pct=0, region=EU, sku_total=992 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "122112", "v2": "12211", "v1": "992", "v3": "121264"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = $504; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 3 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "12", "answer_2": "24", "answer_0": "36", "answer_1": "27"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 91 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "70", "answer_0": "60", "answer_3": "70", "answer_2": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 5 tasks per day for 6 days. Team B finishes 2 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "32", "answer_2": "30", "answer_3": "12", "answer_0": "42"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 7 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p0": "0.500", "p1": "0.500", "p3": "0.462"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = closed; priority = low.\n\nClaim to verify: \"the date is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.667", "p1": "0.300", "p2": "0.300", "p0": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=510, unit_price=768, discount_pct=0, region=EU, sku_total=1278 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "391680", "v2": "39168", "v3": "391170", "v1": "1278"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: date = $711; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "change_address": "Route to the 'change address' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $689. A fee of $183 is charged, then a deposit of $79 arrives. A second fee of $91 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "503", "answer_0": "494", "answer_2": "501", "answer_1": "510"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 3 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_3": "18", "answer_1": "21", "answer_2": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A recipe needs 2 grams of salt per serving and serves 8 people. A shopper bought a 16 gram box and a 160 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.500", "p3": "0.429", "p1": "0.500", "p0": "0.500"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Saturday, December 19, 2026. The SLA requires a response within 13 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = $801; priority = high.\n\nClaim to verify: \"the owner is $801\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: status = $298; priority = low.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Friday, October 02, 2026. The SLA requires a response within 17 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: when will my order arrive\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.636", "p1": "0.636", "p3": "0.300", "p0": "0.364"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 374:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An account starts at $805. A fee of $131 is charged, then a deposit of $46 arrives. A second fee of $65 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "666", "answer_2": "659", "answer_1": "681", "answer_0": "655"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = ops; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Form fields: status = $122; priority = high.\n\nClaim to verify: \"the status is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed October 16, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 20", "date_2": "October 24", "date_3": "October 22", "date_0": "October 22"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket fields: date = team-b; priority = low.\n\nClaim to verify: \"the date is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=225, unit_price=139, discount_pct=0, region=EU, sku_total=364 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "3127", "v3": "31500", "v1": "364", "v0": "31275"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 134 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "150", "answer_1": "156", "answer_3": "139", "answer_2": "146"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A document says the person was born in 1968. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "56", "age_0": "58", "age_1": "57", "age_2": "59"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 6 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you explain this charge on my card\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1966. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "58", "age_2": "59", "age_3": "60", "age_1": "57"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Spreadsheet row: quantity=227, unit_price=265, discount_pct=0, region=EU, sku_total=492 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "492", "v2": "6015", "v3": "59928", "v0": "60155"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.200", "p2": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, correct my address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 770:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed December 09, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 14", "date_1": "December 12", "date_3": "December 14", "date_2": "December 12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Tuesday, January 27, 2026. The SLA requires a response within 3 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 639:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $687. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: date = $983; priority = low.\n\nClaim to verify: \"the date is $983\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: All users logged out, sessions unrecoverable. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A recipe needs 8 grams of salt per serving and serves 6 people. A shopper bought a 48 gram box and a 480 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 5 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "27", "answer_3": "12", "answer_1": "19", "answer_2": "15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1951 per year. It increases by 15% this year, then a $50 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "1658.35", "answer_1": "2243.65", "answer_3": "2225.65", "answer_0": "2193.65"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Payment retries fail for a subset of users. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, January 19, 2026. The SLA requires a response within 5 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: amount = $671; priority = low.\n\nClaim to verify: \"the amount is $671\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you I moved, change the delivery address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p3": "0.167", "p0": "0.333", "p2": "0.167"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, October 23, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed November 07, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 17", "date_0": "November 18", "date_2": "November 20", "date_1": "November 15"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: amount = 2026-11-08; priority = low.\n\nClaim to verify: \"the amount is 2026-11-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.500", "p3": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"excellent service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "praise": "Route to the 'praise' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "40", "answer_1": "19", "answer_3": "28", "answer_2": "12"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $353. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=359, unit_price=655, discount_pct=0, region=EU, sku_total=1014 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "235145", "v3": "234786", "v2": "23514", "v1": "1014"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=522, unit_price=184, discount_pct=0, region=EU, sku_total=706 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "9604", "v0": "96048", "v1": "706", "v3": "95526"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=962, unit_price=560, discount_pct=0, region=EU, sku_total=1522 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "53872", "v1": "1522", "v3": "537758", "v0": "538720"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "play_music": "Route to the 'play music' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "report_damage": "Route to the 'report damage' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p3": "0.167", "p2": "0.500", "p0": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, upgrade my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1982. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "44", "age_3": "42", "age_1": "43", "age_2": "45"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed March 07, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "March 13", "date_2": "March 18", "date_3": "March 15", "date_0": "March 16"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = 2026-09-27; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Team A finishes 4 tasks per day for 4 days. Team B finishes 2 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "16", "answer_0": "24", "answer_3": "8", "answer_1": "18"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 4 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "20", "answer_3": "32", "answer_2": "16", "answer_0": "48"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=851, unit_price=361, discount_pct=0, region=EU, sku_total=1212 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "30721", "v1": "1212", "v3": "308062", "v0": "307211"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed November 17, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "November 24", "date_0": "November 26", "date_3": "November 26", "date_1": "November 24"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Wednesday, August 05, 2026. The SLA requires a response within 10 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.462", "p0": "0.538", "p1": "0.462"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $789. A fee of $104 is charged, then a deposit of $90 arrives. A second fee of $52 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "723", "answer_3": "719", "answer_1": "730", "answer_2": "717"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, December 22, 2026. The SLA requires a response within 14 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 825:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: All users logged out, sessions unrecoverable. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=699, unit_price=131, discount_pct=0, region=EU, sku_total=830 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "830", "v3": "90870", "v0": "91569", "v2": "9156"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "26", "answer_3": "30", "answer_0": "50", "answer_2": "20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 5 people. A shopper bought a 10 gram box and a 100 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Dashboard chart color wrong. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $659 per year. It increases by 25% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "494.25", "answer_3": "750.75", "answer_0": "776.75", "answer_1": "823.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, October 05, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 794:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p0": "0.250", "p2": "0.077"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "Invoice fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A shift runs from 07:15 to 21:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "13h30", "dur_2": "13h45", "dur_3": "13h15", "dur_1": "14h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $655. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket fields: status = ops; priority = high.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1988. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "36", "age_1": "35", "age_3": "38", "age_2": "37"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed June 24, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "June 29", "date_2": "June 27", "date_1": "June 27", "date_0": "June 29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Record fields: amount = $666; priority = high.\n\nClaim to verify: \"the amount is $666\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 4 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 309 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "100", "answer_2": "66", "answer_0": "80", "answer_1": "72"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Monday, July 06, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 6 shifts this week. A client ordered 207 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "84", "answer_2": "97", "answer_1": "68", "answer_3": "103"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Please is my order shipped yet\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 5 shifts this week. A client ordered 340 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "79", "answer_0": "70", "answer_3": "83", "answer_2": "87"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: status = ops; priority = high.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 856:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 689:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you book a table for two tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A document says the person was born in 1973. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "54", "age_0": "53", "age_3": "55", "age_1": "52"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Spreadsheet row: quantity=495, unit_price=315, discount_pct=0, region=EU, sku_total=810 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "155430", "v2": "15592", "v0": "155925", "v1": "810"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 7 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.300", "p3": "0.667", "p2": "0.300", "p0": "0.700"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = 2026-09-27; priority = low.\n\nClaim to verify: \"the amount is 2026-09-27\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=691, unit_price=498, discount_pct=0, region=EU, sku_total=1189 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "34411", "v0": "344118", "v3": "344809", "v1": "1189"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $656. A fee of $139 is charged, then a deposit of $85 arrives. A second fee of $69 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "536", "answer_2": "504", "answer_0": "533", "answer_3": "530"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"update my shipping address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Release checklist for build 407:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: amount = pending; priority = low.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please return my payment\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed April 25, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 04", "date_3": "May 03", "date_1": "May 01", "date_2": "May 02"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.583", "p2": "0.385", "p1": "0.385", "p0": "0.615"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1979. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "45", "age_3": "44", "age_2": "47", "age_0": "46"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $465. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: status = team-b; priority = low.\n\nClaim to verify: \"the status is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = open; priority = low.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed August 13, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 18", "date_2": "August 18", "date_3": "August 20", "date_0": "August 20"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1961. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "63", "age_3": "61", "age_1": "62", "age_2": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Monday, October 12, 2026. The SLA requires a response within 11 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed July 20, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "July 24", "date_3": "July 26", "date_0": "July 24", "date_2": "July 26"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is within the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: owner = $940; priority = high.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Form fields: date = $448; priority = low.\n\nClaim to verify: \"the date is $448\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p1": "0.625", "p3": "0.125", "p2": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $415 per year. It increases by 25% this year, then a $36 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "482.75", "answer_3": "514.75", "answer_2": "311.25", "answer_1": "518.75"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = $416; priority = low.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.333", "p3": "0.083", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: PDF export missing images; copy-paste works. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_subscription": "Route to the 'cancel subscription' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 3 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=458, unit_price=290, discount_pct=0, region=EU, sku_total=748 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "133278", "v0": "132820", "v1": "748", "v2": "13282"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1971. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "56", "age_1": "54", "age_3": "57", "age_0": "55"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.250", "p0": "0.200", "p2": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p1": "0.500", "p3": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $4k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Wednesday, March 25, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: status = $299; priority = high.\n\nClaim to verify: \"the status is $299\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p3": "0.083", "p1": "0.417", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reset my login\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 9 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 251 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "54", "answer_1": "62", "answer_2": "44", "answer_3": "69"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.375", "p3": "0.125", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Tuesday, August 11, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Friday, August 14, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 731:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: status = open; priority = low.\n\nClaim to verify: \"the status is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: status = 2026-08-24; priority = low.\n\nClaim to verify: \"the status is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 3 grams of salt per serving and serves 6 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1045 per year. It increases by 20% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1225", "answer_1": "1254", "answer_3": "1242", "answer_2": "836"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed September 15, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "September 23", "date_0": "September 25", "date_3": "September 25", "date_2": "September 27"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 2 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = 2026-06-17; priority = low.\n\nClaim to verify: \"the owner is 2026-06-17\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Button misaligned on mobile. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p2": "0.100", "p0": "0.150", "p3": "0.050"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Wednesday, February 04, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Sunday, May 24, 2026. The SLA requires a response within 7 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = 2026-11-14; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 813:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = 2026-10-08; priority = low.\n\nClaim to verify: \"the status is 2026-10-08\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: status = 2026-01-10; priority = high.\n\nClaim to verify: \"the status is 2026-01-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Typo in a settings label. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed December 15, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 22", "date_1": "December 20", "date_0": "December 22", "date_2": "December 20"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1983. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "41", "age_2": "43", "age_0": "42", "age_3": "44"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 5 grams of salt per serving and serves 3 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "set_alarm": "Route to the 'set alarm' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Contract signed September 23, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "October 01", "date_2": "October 07", "date_0": "October 05", "date_3": "October 03"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Team A finishes 3 tasks per day for 4 days. Team B finishes 6 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "36", "answer_1": "18", "answer_2": "12", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: owner = $127; priority = high.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "An account starts at $240. A fee of $95 is charged, then a deposit of $51 arrives. A second fee of $47 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "143", "answer_2": "126", "answer_3": "170", "answer_0": "149"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A ticket was opened on Sunday, December 13, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday", "Sunday": "SLA expires on a Sunday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incident report: Backup job delayed but completes. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=647, unit_price=587, discount_pct=0, region=EU, sku_total=1234 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1234", "v2": "37978", "v0": "379789", "v3": "380436"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_order": "Route to the 'cancel order' handler", "change_address": "Route to the 'change address' handler", "praise": "Route to the 'praise' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 4 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.333", "p1": "0.333", "p3": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 7 grams of salt per serving and serves 4 people. A shopper bought a 28 gram box and a 280 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = 2026-03-16; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: status = $324; priority = high.\n\nClaim to verify: \"the status is $324\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 4 shifts this week. A client ordered 213 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "71", "answer_3": "65", "answer_0": "56", "answer_2": "46"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 2 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.714", "p1": "0.714", "p0": "0.286", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = 2026-07-10; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.050", "p2": "0.150", "p1": "0.250", "p0": "0.200"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Release checklist for build 360:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: owner = $861; priority = high.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=304, unit_price=540, discount_pct=0, region=EU, sku_total=844 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "16416", "v0": "164160", "v1": "844", "v3": "164464"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"find me flights to Miami\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "book_flight": "Route to the 'book flight' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Docs page 404s. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p1": "0.019", "p0": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1974. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "52", "age_3": "53", "age_0": "51", "age_1": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=366, unit_price=797, discount_pct=0, region=EU, sku_total=1163 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "291702", "v3": "292068", "v1": "1163", "v2": "29170"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed October 27, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "November 05", "date_2": "November 03", "date_0": "November 05", "date_1": "November 03"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 4 tasks per day for 5 days. Team B finishes 6 tasks per day for 5 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "26", "answer_3": "30", "answer_2": "20", "answer_0": "50"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 5 grams of salt per serving and serves 3 people. A shopper bought a 15 gram box and a 150 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $498. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=577, unit_price=267, discount_pct=0, region=EU, sku_total=844 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "844", "v0": "154059", "v2": "15405", "v3": "154636"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed November 28, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_2": "December 06", "date_1": "December 03", "date_0": "December 04", "date_3": "December 05"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Customer data corrupted in production. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 2 tasks per day for 8 days. Team B finishes 5 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "21", "answer_3": "40", "answer_2": "16", "answer_0": "56"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Spreadsheet row: quantity=966, unit_price=205, discount_pct=0, region=EU, sku_total=1171 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "198996", "v1": "1171", "v0": "198030", "v2": "19803"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_trap"}
{"state": "A subscription costs $1530 per year. It increases by 10% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1666", "answer_2": "1377", "answer_0": "1636", "answer_1": "1683"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = pending; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=380, unit_price=385, discount_pct=0, region=EU, sku_total=765 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "146300", "v3": "146680", "v1": "765", "v2": "14630"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Team A finishes 5 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "64", "answer_3": "24", "answer_1": "43", "answer_2": "40"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=461, unit_price=359, discount_pct=0, region=EU, sku_total=820 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v3": "165038", "v2": "16549", "v0": "165499", "v1": "820"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 28 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Release checklist for build 510:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Customer data corrupted in production. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: amount = 2026-12-01; priority = high.\n\nClaim to verify: \"the amount is 2026-12-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = $565; priority = high.\n\nClaim to verify: \"the date is $565\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 8 hours per shift and runs 5 shifts this week. A client ordered 267 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "128", "answer_3": "114", "answer_0": "120", "answer_2": "122"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Contract signed January 02, 2026. Payment is due within 8 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "January 10", "date_3": "January 12", "date_2": "January 12", "date_0": "January 14"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A shift runs from 11:30 to 17:00 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_2": "5h15", "dur_3": "4h30", "dur_0": "5h00", "dur_1": "5h30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $830 per year. It increases by 25% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "998.5", "answer_3": "971.5", "answer_2": "622.5", "answer_1": "1037.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "63", "answer_2": "21", "answer_1": "27", "answer_3": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, send an sms to my boss\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "weather": "Route to the 'weather' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "book_flight": "Route to the 'book flight' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1970. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "55", "age_0": "54", "age_3": "56", "age_1": "53"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "set_alarm": "Route to the 'set alarm' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "book_flight": "Route to the 'book flight' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "send_message": "Route to the 'send message' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, nothing happens when I click\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p0": "0.250", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 189 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "49", "answer_1": "48", "answer_0": "42", "answer_2": "38"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "An urn holds 8 red and 3 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.727", "p2": "0.273", "p1": "0.273", "p3": "0.700"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: owner = ops; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Release checklist for build 861:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = 2026-01-03; priority = high.\n\nClaim to verify: \"the status is 2026-01-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Tuesday, August 25, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A shift runs from 14:45 to 18:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_1": "3h45", "dur_0": "3h00", "dur_3": "2h45", "dur_2": "3h15"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "technical_issue": "Route to the 'technical issue' handler", "complaint": "Route to the 'complaint' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: date = ops; priority = low.\n\nClaim to verify: \"the date is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.200", "p3": "0.050", "p1": "0.250", "p2": "0.150"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: undo my purchase\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.100", "p2": "0.050", "p1": "0.150", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A shift runs from 12:00 to 12:30 with a 45-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h30", "dur_0": "-1h45", "dur_1": "0h30", "dur_2": "0h00"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 3 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: date = 2026-02-11; priority = high.\n\nClaim to verify: \"the date is 2026-02-11\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 340:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "send_message": "Route to the 'send message' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: Payment retries fail for a subset of users. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Tuesday, April 28, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Friday": "SLA expires on a Friday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 631:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1969. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "55", "age_2": "56", "age_1": "54", "age_3": "53"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"cancel my order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: date = closed; priority = high.\n\nClaim to verify: \"the date is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Record fields: owner = $509; priority = low.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the purchase was a gift. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you wake me at 7\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "report_damage": "Route to the 'report damage' handler", "refund": "Route to the 'refund' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.083", "p0": "0.333", "p2": "0.250", "p1": "0.417"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 287:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1966. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "60", "age_0": "59", "age_3": "61", "age_1": "58"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Form fields: status = ops; priority = low.\n\nClaim to verify: \"the status is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 203:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $265. A fee of $78 is charged, then a deposit of $29 arrives. A second fee of $39 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "177", "answer_2": "159", "answer_3": "151", "answer_1": "191"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed July 13, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 18", "date_0": "July 16", "date_2": "July 18", "date_1": "July 16"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Invoice fields: owner = 2026-02-15; priority = high.\n\nClaim to verify: \"the owner is 2026-02-15\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An account starts at $333. A fee of $61 is charged, then a deposit of $55 arrives. A second fee of $30 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "297", "answer_1": "273", "answer_3": "316", "answer_2": "278"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.150", "p1": "0.250", "p0": "0.200", "p3": "0.050"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 253:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = $835; priority = high.\n\nClaim to verify: \"the amount is $835\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: the app keeps crashing\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "reset_password": "Route to the 'reset password' handler", "complaint": "Route to the 'complaint' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Release checklist for build 347:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Payment retries fail for a subset of users. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.400", "p1": "0.600", "p2": "0.600", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Invoice fields: date = team-a; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A ticket was opened on Sunday, June 07, 2026. The SLA requires a response within 8 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An urn holds 7 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.364", "p2": "0.364", "p0": "0.636", "p3": "0.600"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Record fields: owner = $344; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.462", "p2": "0.462", "p3": "0.500", "p0": "0.538"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A subscription costs $406 per year. It increases by 25% this year, then a $25 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "507.5", "answer_0": "482.5", "answer_3": "515.5", "answer_2": "304.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A warehouse packs 3 crates per hour. It operates 6 hours per shift and runs 3 shifts this week. A client ordered 333 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "40", "answer_0": "54", "answer_1": "48", "answer_2": "68"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A ticket was opened on Wednesday, October 21, 2026. The SLA requires a response within 6 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Saturday": "SLA expires on a Saturday", "Monday": "SLA expires on a Monday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Friday": "SLA expires on a Friday", "Thursday": "SLA expires on a Thursday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A ticket was opened on Friday, September 25, 2026. The SLA requires a response within 20 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Sunday": "SLA expires on a Sunday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 4 days. Team B finishes 7 tasks per day for 4 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "24", "answer_0": "52", "answer_3": "28", "answer_1": "31"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $20k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: $153. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.625", "p3": "0.286", "p2": "0.625", "p0": "0.375"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.625", "p3": "0.286", "p0": "0.375", "p1": "0.625"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A ticket was opened on Friday, March 13, 2026. The SLA requires a response within 15 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Monday": "SLA expires on a Monday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Wednesday": "SLA expires on a Wednesday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.083", "p1": "0.250", "p0": "0.167", "p3": "0.083"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1990. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "34", "age_1": "33", "age_3": "36", "age_2": "35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 5 shifts this week. A client ordered 277 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "75", "answer_2": "94", "answer_3": "69", "answer_1": "87"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=442, unit_price=513, discount_pct=0, region=EU, sku_total=955 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "226746", "v2": "22674", "v1": "955", "v3": "227188"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1975. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "48", "age_0": "49", "age_3": "47", "age_2": "50"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 182 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "87", "answer_2": "107", "answer_3": "72", "answer_0": "90"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 287 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "262", "answer_2": "265", "answer_3": "289", "answer_0": "270"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: status of my delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "track_order": "Route to the 'track order' handler", "cancel_order": "Route to the 'cancel order' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A document says the person was born in 1958. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "68", "age_0": "67", "age_3": "65", "age_1": "66"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "An account starts at $310. A fee of $193 is charged, then a deposit of $87 arrives. A second fee of $96 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "106", "answer_1": "121", "answer_0": "108", "answer_3": "113"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you please cancel order\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "reset_password": "Route to the 'reset password' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 4 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.667", "p1": "0.667", "p3": "0.273", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A document says the person was born in 1977. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "48", "age_2": "50", "age_0": "49", "age_3": "51"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "An account starts at $651. A fee of $108 is charged, then a deposit of $48 arrives. A second fee of $54 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "510", "answer_3": "543", "answer_0": "537", "answer_1": "513"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $501. A fee of $155 is charged, then a deposit of $44 arrives. A second fee of $77 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "339", "answer_0": "313", "answer_1": "288", "answer_2": "324"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"upgrade my subscription\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A recipe needs 9 grams of salt per serving and serves 7 people. A shopper bought a 63 gram box and a 630 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"book a restaurant\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A subscription costs $543 per year. It increases by 10% this year, then a $42 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "555.3", "answer_3": "588.3", "answer_1": "597.3", "answer_2": "488.7"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: amount = team-a; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.333", "p1": "0.500", "p3": "0.167", "p2": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Record fields: owner = open; priority = low.\n\nClaim to verify: \"the owner is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1986. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "38", "age_1": "37", "age_2": "39", "age_3": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "An account starts at $756. A fee of $147 is charged, then a deposit of $82 arrives. A second fee of $73 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "618", "answer_3": "616", "answer_1": "637", "answer_2": "645"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"reroute to a different address\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "praise": "Route to the 'praise' handler", "change_address": "Route to the 'change address' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - I want to cancel the order I just placed\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $262. A fee of $114 is charged, then a deposit of $41 arrives. A second fee of $57 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "149", "answer_2": "140", "answer_0": "132", "answer_1": "145"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Release checklist for build 474:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"website wont load\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "technical_issue": "Route to the 'technical issue' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Button misaligned on mobile. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: amount = 2026-03-05; priority = low.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "send_message": "Route to the 'send message' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "A ticket was opened on Thursday, June 11, 2026. The SLA requires a response within 4 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Tuesday": "SLA expires on a Tuesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $580. A fee of $115 is charged, then a deposit of $31 arrives. A second fee of $57 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "439", "answer_3": "417", "answer_2": "416", "answer_1": "447"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"weather": "Route to the 'weather' handler", "track_order": "Route to the 'track order' handler", "praise": "Route to the 'praise' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "report_damage": "Route to the 'report damage' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: is it going to rain\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "report_damage": "Route to the 'report damage' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = 2026-08-19; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = $228; priority = high.\n\nClaim to verify: \"the date is $228\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A document says the person was born in 1984. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "40", "age_2": "42", "age_0": "41", "age_3": "39"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A shift runs from 21:00 to 21:45 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_3": "-1h45", "dur_2": "0h45", "dur_0": "0h15", "dur_1": "0h45"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Contract signed April 14, 2026. Payment is due within 7 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "April 23", "date_0": "April 23", "date_2": "April 25", "date_1": "April 21"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1962. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_0": "64", "age_1": "63", "age_2": "65", "age_3": "66"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 4 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "36", "answer_1": "40", "answer_0": "60", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1012 per year. It increases by 15% this year, then a $31 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "860.2", "answer_1": "1163.8", "answer_0": "1132.8", "answer_3": "1110.8"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "An account starts at $381. A fee of $164 is charged, then a deposit of $45 arrives. A second fee of $82 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "180", "answer_1": "183", "answer_2": "204", "answer_3": "187"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Spreadsheet row: quantity=927, unit_price=455, discount_pct=0, region=EU, sku_total=1382 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "421785", "v3": "422712", "v1": "1382", "v2": "42178"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"Please will it snow tonight\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"restaurant_reservation": "Route to the 'restaurant reservation' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler", "book_flight": "Route to the 'book flight' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Docs page 404s. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $393. A fee of $198 is charged, then a deposit of $82 arrives. A second fee of $99 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "178", "answer_3": "154", "answer_2": "162", "answer_1": "168"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Login fails for users on one provider. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Form fields: owner = $54; priority = high.\n\nClaim to verify: \"the owner is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed January 20, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "January 28", "date_1": "January 26", "date_2": "January 30", "date_0": "January 28"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 8 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.667", "p2": "0.333", "p1": "0.333", "p3": "0.636"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I want the premium tier\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "billing_question": "Route to the 'billing question' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "cancel_order": "Route to the 'cancel order' handler", "weather": "Route to the 'weather' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "reset_password": "Route to the 'reset password' handler", "weather": "Route to the 'weather' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"table for 7pm\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"change_address": "Route to the 'change address' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "report_damage": "Route to the 'report damage' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A ticket was opened on Tuesday, February 03, 2026. The SLA requires a response within 12 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Friday": "SLA expires on a Friday", "Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p3": "0.500", "p2": "0.077", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Record fields: status = 2026-08-25; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p3": "0.500", "p0": "0.250"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1971. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "52", "age_3": "51", "age_2": "54", "age_0": "53"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Spreadsheet row: quantity=412, unit_price=133, discount_pct=0, region=EU, sku_total=545 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "5479", "v3": "54384", "v1": "545", "v0": "54796"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Monday, November 23, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Tuesday": "SLA expires on a Tuesday", "Saturday": "SLA expires on a Saturday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Wednesday": "SLA expires on a Wednesday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1949 per year. It increases by 15% this year, then a $39 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "2202.35", "answer_2": "1656.65", "answer_3": "2220.35", "answer_1": "2241.35"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 6 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.455", "p1": "0.500", "p0": "0.500", "p2": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A shift runs from 08:30 to 15:30 with a 30-minute unpaid break.", "questions": {"q0": {"type": "choice", "instructions": "Choose the paid duration.", "criteria": {"dur_0": "6h30", "dur_3": "6h00", "dur_1": "7h00", "dur_2": "7h00"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1378 per year. It increases by 20% this year, then a $44 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1609.6", "answer_3": "1641.6", "answer_1": "1653.6", "answer_2": "1102.4"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.375", "p3": "0.125", "p1": "0.500", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - book airfare\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket fields: owner = 2026-07-10; priority = high.\n\nClaim to verify: \"the owner is 2026-07-10\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.333", "p2": "0.500", "p1": "0.500", "p0": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Release checklist for build 260:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 8 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.727", "p2": "0.727", "p0": "0.273", "p3": "0.200"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Please it came cracked\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "complaint": "Route to the 'complaint' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = $392; priority = low.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A warehouse packs 4 crates per hour. It operates 2 hours per shift and runs 2 shifts this week. A client ordered 363 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "16", "answer_3": "21", "answer_2": "35", "answer_1": "10"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Login fails for users on one provider. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $281. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p2": "0.077", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - great job on the delivery\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "praise": "Route to the 'praise' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: status = closed; priority = high.\n\nClaim to verify: \"the status is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = 2026-02-03; priority = low.\n\nClaim to verify: \"the amount is 2026-02-03\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Form fields: amount = pending; priority = high.\n\nClaim to verify: \"the amount is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Backup job delayed but completes. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Typo in a settings label. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 8 hours per shift and runs 6 shifts this week. A client ordered 380 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "319", "answer_1": "344", "answer_0": "336", "answer_3": "351"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Team A finishes 3 tasks per day for 6 days. Team B finishes 6 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "54", "answer_3": "36", "answer_1": "24", "answer_2": "18"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: owner = $680; priority = high.\n\nClaim to verify: \"the owner is $680\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Record fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is closed\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "An urn holds 5 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.364", "p2": "0.583", "p1": "0.583", "p0": "0.417"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 6 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 74 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_3": "117", "answer_2": "106", "answer_1": "111"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 4 hours per shift and runs 2 shifts this week. A client ordered 111 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "86", "answer_0": "72", "answer_2": "54", "answer_3": "64"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "An account starts at $730. A fee of $187 is charged, then a deposit of $32 arrives. A second fee of $93 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "459", "answer_1": "489", "answer_3": "497", "answer_0": "482"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"screen is shattered on arrival\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "report_damage": "Route to the 'report damage' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: status = ops; priority = high.\n\nClaim to verify: \"the status is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Login fails for users on one provider. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - what is this billing line\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "reset_password": "Route to the 'reset password' handler", "billing_question": "Route to the 'billing question' handler", "change_address": "Route to the 'change address' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 3 crates per hour. It operates 5 hours per shift and runs 6 shifts this week. A client ordered 243 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "95", "answer_2": "75", "answer_3": "78", "answer_0": "90"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "An urn holds 5 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.286", "p3": "0.667", "p2": "0.286", "p0": "0.714"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 6 grams of salt per serving and serves 12 people. A shopper bought a 72 gram box and a 720 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = ops; priority = low.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A subscription costs $1399 per year. It increases by 5% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1436.95", "answer_3": "1408.95", "answer_2": "1329.05", "answer_1": "1468.95"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 2 grams of salt per serving and serves 10 people. A shopper bought a 20 gram box and a 200 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: Login fails for users on one provider. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket fields: status = team-a; priority = high.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A recipe needs 3 grams of salt per serving and serves 4 people. A shopper bought a 12 gram box and a 120 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed February 28, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 13", "date_1": "March 10", "date_3": "March 12", "date_2": "March 11"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A subscription costs $1237 per year. It increases by 10% this year, then a $48 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "1320.7", "answer_0": "1312.7", "answer_1": "1360.7", "answer_2": "1113.3"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: amount = $300; priority = low.\n\nClaim to verify: \"the amount is $300\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1977. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "46", "age_2": "49", "age_0": "48", "age_1": "47"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p3": "0.500", "p0": "0.250", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Spreadsheet row: quantity=411, unit_price=364, discount_pct=0, region=EU, sku_total=775 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "775", "v0": "149604", "v2": "14960", "v3": "149193"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=929, unit_price=785, discount_pct=0, region=EU, sku_total=1714 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "729265", "v3": "728336", "v2": "72926", "v1": "1714"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Contract signed September 05, 2026. Payment is due within 10 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "September 18", "date_1": "September 15", "date_3": "September 17", "date_2": "September 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket summary: customer requests escalation. The system record shows budget line: $39k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Contract signed March 08, 2026. Payment is due within 6 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "March 16", "date_3": "March 16", "date_2": "March 18", "date_1": "March 14"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "cancel_order": "Route to the 'cancel order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 7 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.778", "p0": "0.222", "p3": "0.125", "p2": "0.778"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was opened software. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Team A finishes 4 tasks per day for 8 days. Team B finishes 6 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "38", "answer_0": "80", "answer_3": "48", "answer_2": "32"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Record fields: amount = 2026-01-24; priority = low.\n\nClaim to verify: \"the amount is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Please password not working\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "reset_password": "Route to the 'reset password' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"play some jazz\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p2": "0.167", "p3": "0.167", "p0": "0.333"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "A fair 12-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p3": "0.083", "p1": "0.333", "p0": "0.250"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - temperature outside\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "cancel_order": "Route to the 'cancel order' handler", "complaint": "Route to the 'complaint' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 17 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 4 grams of salt per serving and serves 11 people. A shopper bought a 44 gram box and a 440 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A ticket was opened on Tuesday, November 10, 2026. The SLA requires a response within 18 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Tuesday": "SLA expires on a Tuesday", "Wednesday": "SLA expires on a Wednesday", "Monday": "SLA expires on a Monday", "Thursday": "SLA expires on a Thursday", "Friday": "SLA expires on a Friday", "Saturday": "SLA expires on a Saturday", "Sunday": "SLA expires on a Sunday"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A subscription costs $963 per year. It increases by 15% this year, then a $14 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1093.45", "answer_1": "1107.45", "answer_3": "1115.45", "answer_2": "818.55"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Form fields: date = ops; priority = high.\n\nClaim to verify: \"the date is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An urn holds 6 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.400", "p3": "0.556", "p2": "0.400", "p0": "0.600"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler", "play_music": "Route to the 'play music' handler", "reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "DATABASE RECORD (verified Apr 2026):\n  subscription_status: active\n  renewal_date: 20 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. The system record shows delivery date: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Button misaligned on mobile. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A recipe needs 7 grams of salt per serving and serves 6 people. A shopper bought a 42 gram box and a 420 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 26 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 6 red and 5 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p1": "0.455", "p2": "0.455", "p0": "0.545"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1984. The record was checked in 2024.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "41", "age_1": "39", "age_0": "40", "age_3": "42"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: no physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 2 grams of salt per serving and serves 9 people. A shopper bought a 18 gram box and a 180 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incident report: Search returns slow results but works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 5 crates per hour. It operates 3 hours per shift and runs 4 shifts this week. A client ordered 50 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "73", "answer_3": "57", "answer_2": "64", "answer_0": "60"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - worst experience ever\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"set_alarm": "Route to the 'set alarm' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "complaint": "Route to the 'complaint' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you play the new album\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "play_music": "Route to the 'play music' handler", "book_flight": "Route to the 'book flight' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Dashboard chart color wrong. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An account starts at $448. A fee of $101 is charged, then a deposit of $88 arrives. A second fee of $50 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "387", "answer_3": "382", "answer_0": "385", "answer_2": "367"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, dinner reservation for four\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "book_flight": "Route to the 'book flight' handler", "send_message": "Route to the 'send message' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A warehouse packs 6 crates per hour. It operates 5 hours per shift and runs 4 shifts this week. A client ordered 313 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "120", "answer_1": "113", "answer_2": "116", "answer_3": "109"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A warehouse packs 9 crates per hour. It operates 3 hours per shift and runs 3 shifts this week. A client ordered 83 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "62", "answer_3": "91", "answer_1": "79", "answer_0": "81"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "book_flight": "Route to the 'book flight' handler", "play_music": "Route to the 'play music' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.019", "p0": "0.250", "p2": "0.077", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 1 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Form fields: amount = open; priority = high.\n\nClaim to verify: \"the amount is open\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Invoice fields: status = 2026-10-07; priority = high.\n\nClaim to verify: \"the status is 2026-10-07\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Release checklist for build 253:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, very happy with the product\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "refund": "Route to the 'refund' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "billing_question": "Route to the 'billing question' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $375. A fee of $56 is charged, then a deposit of $49 arrives. A second fee of $28 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "350", "answer_0": "340", "answer_3": "370", "answer_1": "321"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: Typo in a settings label. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.125", "p2": "0.125", "p1": "0.375"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p1": "0.019", "p3": "0.500", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A document says the person was born in 1967. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_3": "60", "age_1": "57", "age_2": "59", "age_0": "58"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"Please put on my workout playlist\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Record fields: amount = team-a; priority = high.\n\nClaim to verify: \"the amount is team-a\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported late. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 20-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.200", "p2": "0.100", "p3": "0.050", "p0": "0.150"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $657. A fee of $42 is charged, then a deposit of $40 arrives. A second fee of $21 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "650", "answer_3": "651", "answer_0": "634", "answer_2": "612"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Form fields: status = pending; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed November 26, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "December 02", "date_3": "December 02", "date_1": "November 30", "date_2": "November 30"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Form fields: owner = $972; priority = high.\n\nClaim to verify: \"the owner is $972\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"book a flight to Denver\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "praise": "Route to the 'praise' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "book_flight": "Route to the 'book flight' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: PDF export missing images; copy-paste works. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 7 hours per shift and runs 5 shifts this week. A client ordered 397 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "232", "answer_3": "247", "answer_0": "245", "answer_1": "236"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 4 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"help me get back into my account\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"send_message": "Route to the 'send message' handler", "weather": "Route to the 'weather' handler", "refund": "Route to the 'refund' handler", "technical_issue": "Route to the 'technical issue' handler", "reset_password": "Route to the 'reset password' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=575, unit_price=787, discount_pct=0, region=EU, sku_total=1362 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "45252", "v1": "1362", "v0": "452525", "v3": "451950"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item is electronics. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Hey, this is unacceptable\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "complaint": "Route to the 'complaint' handler", "track_order": "Route to the 'track order' handler", "weather": "Route to the 'weather' handler", "change_address": "Route to the 'change address' handler"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Team A finishes 6 tasks per day for 3 days. Team B finishes 4 tasks per day for 3 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "18", "answer_0": "30", "answer_1": "22", "answer_3": "12"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Contract signed August 14, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "August 18", "date_0": "August 20", "date_2": "August 18", "date_3": "August 20"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A document says the person was born in 1970. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "55", "age_0": "56", "age_2": "57", "age_3": "54"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Team A finishes 6 tasks per day for 6 days. Team B finishes 5 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "66", "answer_3": "30", "answer_1": "41", "answer_2": "36"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"set a timer for 20 minutes\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "refund": "Route to the 'refund' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "upgrade_plan": "Route to the 'upgrade plan' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Invoice fields: amount = 2026-10-18; priority = high.\n\nClaim to verify: \"the amount is unknown\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Ticket fields: amount = team-b; priority = high.\n\nClaim to verify: \"the amount is team-b\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by wind in a coastal zone. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Invoice fields: date = 2026-04-09; priority = low.\n\nClaim to verify: \"the date is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"message the group chat\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No manager name is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: $39k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"reset_password": "Route to the 'reset password' handler", "track_order": "Route to the 'track order' handler", "send_message": "Route to the 'send message' handler", "play_music": "Route to the 'play music' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A warehouse packs 6 crates per hour. It operates 6 hours per shift and runs 2 shifts this week. A client ordered 58 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_3": "69", "answer_0": "72", "answer_2": "74", "answer_1": "58"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Invoice fields: status = 2026-11-23; priority = low.\n\nClaim to verify: \"the status is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1968. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "58", "age_3": "55", "age_1": "56", "age_0": "57"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A recipe needs 2 grams of salt per serving and serves 12 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A recipe needs 3 grams of salt per serving and serves 7 people. A shopper bought a 21 gram box and a 210 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: PDF export missing images; copy-paste works. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "An urn holds 2 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.500", "p3": "0.333", "p2": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 30 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"terrible support\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"track_order": "Route to the 'track order' handler", "complaint": "Route to the 'complaint' handler", "set_alarm": "Route to the 'set alarm' handler", "billing_question": "Route to the 'billing question' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.667", "p0": "0.500", "p2": "0.333", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incident report: Customer data corrupted in production. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Incident report: Search returns slow results but works. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"refund": "Route to the 'refund' handler", "send_message": "Route to the 'send message' handler", "report_damage": "Route to the 'report damage' handler", "technical_issue": "Route to the 'technical issue' handler", "play_music": "Route to the 'play music' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "A recipe needs 6 grams of salt per serving and serves 6 people. A shopper bought a 36 gram box and a 360 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=848, unit_price=827, discount_pct=0, region=EU, sku_total=1675 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "701296", "v1": "1675", "v2": "70129", "v3": "702144"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Spreadsheet row: quantity=956, unit_price=102, discount_pct=0, region=EU, sku_total=1058 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "97512", "v3": "98468", "v2": "9751", "v1": "1058"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "A fair 8-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.500", "p0": "0.375", "p3": "0.125", "p2": "0.250"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A warehouse packs 3 crates per hour. It operates 4 hours per shift and runs 5 shifts this week. A client ordered 228 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "60", "answer_1": "57", "answer_2": "68", "answer_3": "46"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket fields: status = $428; priority = high.\n\nClaim to verify: \"the status is $428\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"end my membership\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "track_order": "Route to the 'track order' handler", "report_damage": "Route to the 'report damage' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Invoice fields: amount = $722; priority = high.\n\nClaim to verify: \"the amount is $9999\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Invoice fields: amount = 2026-06-04; priority = low.\n\nClaim to verify: \"the amount is 2026-06-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incident report: API returns 500s for one endpoint. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Invoice fields: owner = ops; priority = high.\n\nClaim to verify: \"the owner is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"love the new update\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "play_music": "Route to the 'play music' handler", "send_message": "Route to the 'send message' handler", "praise": "Route to the 'praise' handler", "complaint": "Route to the 'complaint' handler", "weather": "Route to the 'weather' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: Backup job delayed but completes. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A warehouse packs 7 crates per hour. It operates 2 hours per shift and runs 3 shifts this week. A client ordered 391 crates.\n\nHow many crates does the warehouse pack this week?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "42", "answer_1": "55", "answer_3": "38", "answer_2": "29"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: $33k. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 3 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.500", "p2": "0.333", "p1": "0.667", "p3": "0.167"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Team A finishes 2 tasks per day for 6 days. Team B finishes 7 tasks per day for 6 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "19", "answer_2": "12", "answer_3": "42", "answer_0": "54"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "An account starts at $553. A fee of $119 is charged, then a deposit of $84 arrives. A second fee of $59 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "465", "answer_0": "459", "answer_2": "481", "answer_3": "470"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $220 per year. It increases by 20% this year, then a $24 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "264", "answer_2": "176", "answer_3": "257", "answer_0": "240"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A subscription costs $1462 per year. It increases by 5% this year, then a $47 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1488.1", "answer_3": "1469.1", "answer_1": "1535.1", "answer_2": "1388.9"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No contract term is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: date = $232; priority = low.\n\nClaim to verify: \"the date is $232\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.077", "p1": "0.019", "p0": "0.250", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $1691 per year. It increases by 10% this year, then a $32 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1828.1", "answer_1": "1860.1", "answer_2": "1521.9", "answer_3": "1799.1"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = $385; priority = low.\n\nClaim to verify: \"the amount is $385\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p2": "0.077", "p1": "0.019", "p3": "0.500"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"I need help: set an alarm for tomorrow morning\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "set_alarm": "Route to the 'set alarm' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "billing_question": "Route to the 'billing question' handler", "praise": "Route to the 'praise' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Form fields: owner = 2026-08-21; priority = low.\n\nClaim to verify: \"the owner is 2026-08-21\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Contract signed May 12, 2026. Payment is due within 5 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 19", "date_1": "May 17", "date_0": "May 19", "date_2": "May 21"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "Contract signed January 28, 2026. Payment is due within 4 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_1": "February 01", "date_2": "February 05", "date_0": "February 03", "date_3": "February 03"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incoming user message to the assistant router:\n\n\"alarm at 6:30 please\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "change_address": "Route to the 'change address' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. No refund amount is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: owner = pending; priority = high.\n\nClaim to verify: \"the owner is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"play music\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"upgrade_plan": "Route to the 'upgrade plan' handler", "refund": "Route to the 'refund' handler", "billing_question": "Route to the 'billing question' handler", "track_order": "Route to the 'track order' handler", "play_music": "Route to the 'play music' handler", "weather": "Route to the 'weather' handler", "set_alarm": "Route to the 'set alarm' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Incident report: API returns 500s for one endpoint. It appeared after the last deploy.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_ordinal"}
{"state": "A ticket was opened on Sunday, September 13, 2026. The SLA requires a response within 19 calendar days.\n\nOn which weekday does the SLA expire?", "questions": {"q0": {"type": "choice", "instructions": "Choose the weekday on which the SLA expires.", "criteria": {"Monday": "SLA expires on a Monday", "Friday": "SLA expires on a Friday", "Sunday": "SLA expires on a Sunday", "Thursday": "SLA expires on a Thursday", "Saturday": "SLA expires on a Saturday", "Wednesday": "SLA expires on a Wednesday", "Tuesday": "SLA expires on a Tuesday"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"stop the monthly charges\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"complaint": "Route to the 'complaint' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "reset_password": "Route to the 'reset password' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: a physician note was provided. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed May 02, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "May 07", "date_1": "May 05", "date_0": "May 06", "date_2": "May 04"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_temporal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows refund amount: 12 days. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the refund amount? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Ticket summary: customer requests escalation. No budget line is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the budget line? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 2 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p2": "0.167", "p0": "0.333", "p3": "0.167", "p1": "0.500"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $10000 per occurrence.\n  (4) All claims carry a $500 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the damage was caused by flooding. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Record fields: amount = closed; priority = high.\n\nClaim to verify: \"the amount is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 60 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item was marked final sale. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "A recipe needs 3 grams of salt per serving and serves 11 people. A shopper bought a 33 gram box and a 330 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: owner = closed; priority = low.\n\nClaim to verify: \"the owner is deleted\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Contract signed May 14, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_0": "May 27", "date_1": "May 23", "date_2": "May 29", "date_3": "May 25"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 8 months tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Contract signed December 01, 2026. Payment is due within 3 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "December 06", "date_2": "December 02", "date_0": "December 04", "date_1": "December 04"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Incident report: Docs page 404s. Users reported it during peak hours.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $2000 per occurrence.\n  (4) All claims carry a $250 deductible.\n  (5) Claims reported within 14 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the loss is a theft. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Incident report: All users logged out, sessions unrecoverable. A customer mentioned it in a survey.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_ordinal"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: R. Alvarez. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Spreadsheet row: quantity=660, unit_price=235, discount_pct=0, region=EU, sku_total=895 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v0": "155100", "v1": "895", "v2": "15510", "v3": "155760"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Record fields: status = pending; priority = low.\n\nClaim to verify: \"the status is pending\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=424, unit_price=533, discount_pct=0, region=EU, sku_total=957 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v2": "22599", "v1": "957", "v3": "226416", "v0": "225992"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "Invoice fields: status = 2026-08-04; priority = low.\n\nClaim to verify: \"the status is 2026-08-04\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 8 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request is for parental leave as primary caregiver. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A subscription costs $716 per year. It increases by 5% this year, then a $10 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "680.2", "answer_0": "741.8", "answer_3": "721.8", "answer_1": "751.8"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"I am very unhappy with your service\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"report_damage": "Route to the 'report damage' handler", "complaint": "Route to the 'complaint' handler", "change_address": "Route to the 'change address' handler", "play_music": "Route to the 'play music' handler", "praise": "Route to the 'praise' handler", "book_flight": "Route to the 'book flight' handler", "cancel_subscription": "Route to the 'cancel subscription' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An account starts at $521. A fee of $169 is charged, then a deposit of $60 arrives. A second fee of $84 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "335", "answer_3": "343", "answer_0": "328", "answer_1": "311"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "Incoming user message to the assistant router:\n\n\"Can you forecast for the weekend\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "track_order": "Route to the 'track order' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "complaint": "Route to the 'complaint' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_intent"}
{"state": "An account starts at $501. A fee of $124 is charged, then a deposit of $27 arrives. A second fee of $62 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "328", "answer_3": "370", "answer_0": "342", "answer_1": "320"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A subscription costs $1534 per year. It increases by 25% this year, then a $29 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_0": "1888.5", "answer_3": "1899.5", "answer_1": "1917.5", "answer_2": "1150.5"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "A recipe needs 7 grams of salt per serving and serves 11 people. A shopper bought a 77 gram box and a 770 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "POLICY (INSURANCE)\n  (1) Flood damage is excluded under this policy.\n  (2) Wind damage is excluded in coastal zones.\n  (3) Theft is covered up to $5000 per occurrence.\n  (4) All claims carry a $1000 deductible.\n  (5) Claims reported within 30 days of loss qualify for expedited handling.\n\nCLAIM FACTS\nReviewer's note: the claim was reported promptly. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"cover_minus_deductible": "Outcome 'cover minus deductible' applies", "deny_excluded_peril": "Outcome 'deny excluded peril' applies", "deny_late_report": "Outcome 'deny late report' applies", "cover_expedited": "Outcome 'cover expedited' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "A recipe needs 6 grams of salt per serving and serves 4 people. A shopper bought a 24 gram box and a 240 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified Mar 2026):\n  subscription_status: active\n  renewal_date: 7 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Incoming user message to the assistant router:\n\n\"reserve a spot at the italian place\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"cancel_order": "Route to the 'cancel order' handler", "restaurant_reservation": "Route to the 'restaurant reservation' handler", "upgrade_plan": "Route to the 'upgrade plan' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 20% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the item arrived defective. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "Release checklist for build 834:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A document says the person was born in 1990. The record was checked in 2026.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_1": "35", "age_3": "38", "age_2": "37", "age_0": "36"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.500", "p0": "0.250", "p2": "0.077", "p1": "0.019"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_probability"}
{"state": "An urn holds 4 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.600", "p0": "0.400", "p3": "0.333", "p2": "0.600"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An urn holds 3 red and 6 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p3": "0.250", "p2": "0.667", "p0": "0.333", "p1": "0.667"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "An account starts at $897. A fee of $40 is charged, then a deposit of $25 arrives. A second fee of $20 is charged.\n\nWhat is the final balance?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "872", "answer_0": "862", "answer_3": "837", "answer_1": "832"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0]}, "family": "jb_multihop"}
{"state": "A standard 52-card deck; one card is drawn.\n\nWhat is the probability the card is a heart?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.250", "p3": "0.500", "p1": "0.019", "p2": "0.077"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "A recipe needs 2 grams of salt per serving and serves 3 people. A shopper bought a 6 gram box and a 60 gram refill pack.\n\nDoes one box suffice for one batch of the recipe?", "questions": {"q0": {"type": "choice", "instructions": "Compute servings x grams; compare to the box size, not the refill pack.", "criteria": {"yes_exactly": "Yes, exactly enough", "yes_spare": "Yes, with spare", "no": "No, insufficient"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "DATABASE RECORD (verified May 2026):\n  subscription_status: active\n  renewal_date: 19 Jun 2026\n\nUNVERIFIED STICKY NOTE (dated 2024):\n  \"customer cancelled - account closed\"\n\nIs the subscription active per the verified record?", "questions": {"q0": {"type": "choice", "instructions": "Answer using only the verified record; the sticky note is stale.", "criteria": {"active": "Subscription is active", "cancelled": "Subscription is cancelled", "unknown": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "An urn holds 3 red and 4 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p0": "0.429", "p1": "0.571", "p2": "0.571", "p3": "0.333"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "Ticket summary: customer requests escalation. The system record shows contract term: 24 months. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the contract term? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Form fields: amount = ops; priority = high.\n\nClaim to verify: \"the amount is ops\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "A document says the person was born in 1980. The record was checked in 2025.\n\nHow old was the person when checked? (Assume birthday already occurred.)", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct age.", "criteria": {"age_2": "46", "age_3": "43", "age_0": "45", "age_1": "44"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Ticket fields: owner = 2026-06-06; priority = low.\n\nClaim to verify: \"the owner is 2026-06-06\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 14 days only if defective.\n  (4) Electronics returns incur a 15% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Ticket summary: customer requests escalation. The system record shows manager name: $701. No other fields populated.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the manager name? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [1.0, 0.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Incident report: Dashboard chart color wrong. One engineer could not reproduce it.", "questions": {"q0": {"type": "choice", "instructions": "Rate severity by the strict rubric; the highest level that the facts support.", "criteria": {"sev_0": "No impact; cosmetic or documentation only", "sev_1": "Minor impairment; workaround exists", "sev_2": "Major impairment; core function degraded", "sev_3": "Critical; outage, data loss, or safety risk"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_ordinal"}
{"state": "Team A finishes 6 tasks per day for 8 days. Team B finishes 3 tasks per day for 8 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "48", "answer_1": "51", "answer_0": "72", "answer_3": "24"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 8 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the request would exceed the annual cap. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 0.0, 1.0]}, "family": "jb_policy"}
{"state": "Ticket fields: owner = team-b; priority = low.\n\nClaim to verify: \"the owner is 1970-01-01\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 0.0}, "family": "jb_extract"}
{"state": "Incoming user message to the assistant router:\n\n\"Hi - start the radio\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "play_music": "Route to the 'play music' handler", "technical_issue": "Route to the 'technical issue' handler"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_intent"}
{"state": "Contract signed July 20, 2026. Payment is due within 9 business days (Mon-Fri only, ignoring holidays).\n\nWhat is the payment due date?", "questions": {"q0": {"type": "choice", "instructions": "Choose the correct due date.", "criteria": {"date_3": "July 31", "date_2": "July 29", "date_0": "July 31", "date_1": "July 29"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_temporal"}
{"state": "Release checklist for build 973:\n  [x] unit tests pass\n  [x] docs updated\n  [ ] security scan clean\n  [x] changelog written\n\nPolicy: release is blocked only if a REQUIRED item is unchecked. All four items are required.", "questions": {"q0": {"type": "choice", "instructions": "Apply the policy literally: any unchecked required item blocks the release.", "criteria": {"release_ok": "Release may proceed", "blocked_security": "Blocked: security scan not complete", "blocked_docs": "Blocked: docs missing"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_trap"}
{"state": "A subscription costs $305 per year. It increases by 10% this year, then a $12 loyalty credit is applied.\n\nWhat is the effective yearly cost?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_2": "274.5", "answer_0": "323.5", "answer_1": "335.5", "answer_3": "291.5"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_multihop"}
{"state": "POLICY (LEAVE)\n  (1) Employees with <1 year tenure get no extended leave.\n  (2) Parental leave: 12 weeks paid for primary caregivers.\n  (3) Medical leave requires a physician note.\n  (4) Total paid leave cannot exceed 6 weeks per year.\n\nCLAIM FACTS\nReviewer's note: the employee has 3 years tenure. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"approve_paid": "Outcome 'approve paid' applies", "approve_unpaid": "Outcome 'approve unpaid' applies", "deny_no_documentation": "Outcome 'deny no documentation' applies", "deny_tenure": "Outcome 'deny tenure' applies", "deny_cap_exceeded": "Outcome 'deny cap exceeded' applies"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0]}, "family": "jb_policy"}
{"state": "A fair 6-sided die is rolled once.\n\nWhat is the probability the roll is 4 or less?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.833", "p2": "0.500", "p3": "0.167", "p0": "0.667"}}}, "targets": {"q0": [0.0, 0.0, 1.0, 0.0]}, "family": "jb_probability"}
{"state": "Incoming user message to the assistant router:\n\n\"getting an error message\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"technical_issue": "Route to the 'technical issue' handler", "cancel_order": "Route to the 'cancel order' handler", "refund": "Route to the 'refund' handler", "change_address": "Route to the 'change address' handler", "weather": "Route to the 'weather' handler"}}}, "targets": {"q0": [1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "An urn holds 7 red and 2 blue marbles. One marble is drawn uniformly at random.\n\nWhat is the probability the marble is red?", "questions": {"q0": {"type": "choice", "instructions": "Choose the exact probability.", "criteria": {"p1": "0.222", "p3": "0.750", "p2": "0.222", "p0": "0.778"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_probability"}
{"state": "POLICY (REFUND)\n  (1) Standard items may be returned within 14 days for a full refund.\n  (2) Items marked 'final sale' cannot be returned or refunded.\n  (3) Opened software and media may be returned within 7 days only if defective.\n  (4) Electronics returns incur a 10% restocking fee.\n  (5) Gift purchases are refunded as store credit only.\n\nCLAIM FACTS\nReviewer's note: the return is past the allowed window. All other facts are standard and unremarkable.\n\nA junior adjuster wrote: \"surface answer looks like a routine approval\".", "questions": {"q0": {"type": "choice", "instructions": "As the adjudicator, apply the policy clauses to the facts. The correct outcome follows the controlling clause, not the surface impression.", "criteria": {"full_refund": "Outcome 'full refund' applies", "store_credit": "Outcome 'store credit' applies", "defective_exchange": "Outcome 'defective exchange' applies", "deny_return": "Outcome 'deny return' applies", "refund_minus_restocking": "Outcome 'refund minus restocking' applies"}}}, "targets": {"q0": [0.0, 0.0, 0.0, 1.0, 0.0]}, "family": "jb_policy"}
{"state": "Incoming user message to the assistant router:\n\n\"you billed me wrong\"", "questions": {"q0": {"type": "choice", "instructions": "Pick the single intent this message maps to.", "criteria": {"book_flight": "Route to the 'book flight' handler", "billing_question": "Route to the 'billing question' handler", "set_alarm": "Route to the 'set alarm' handler", "cancel_subscription": "Route to the 'cancel subscription' handler", "report_damage": "Route to the 'report damage' handler", "track_order": "Route to the 'track order' handler"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0, 0.0, 0.0]}, "family": "jb_intent"}
{"state": "Ticket summary: customer requests escalation. No delivery date is recorded anywhere in the record. Other fields are present.", "questions": {"q0": {"type": "choice", "instructions": "Does the record state the delivery date? Answer only what the record contains.", "criteria": {"present": "The field is recorded", "absent": "The field is not recorded", "unclear": "Cannot be determined"}}}, "targets": {"q0": [0.0, 1.0, 0.0]}, "family": "jb_ambiguity"}
{"state": "Record fields: status = 2026-01-09; priority = high.\n\nClaim to verify: \"the status is 2026-01-09\".", "questions": {"q0": {"type": "noul", "instructions": "Decide whether the claim is supported by the fields shown."}}, "targets": {"q0": 1.0}, "family": "jb_extract"}
{"state": "Spreadsheet row: quantity=967, unit_price=648, discount_pct=0, region=EU, sku_total=1615 (this column holds quantity + unit_price, NOT revenue).\n\nWhich value equals revenue (quantity x unit_price)?", "questions": {"q0": {"type": "choice", "instructions": "Pick the cell semantics correctly; sku_total is a decoy.", "criteria": {"v1": "1615", "v3": "625649", "v2": "62661", "v0": "626616"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_trap"}
{"state": "Team A finishes 2 tasks per day for 7 days. Team B finishes 6 tasks per day for 7 days.\n\nHow many tasks did both teams finish in total?", "questions": {"q0": {"type": "choice", "instructions": "Compute the answer. Only the exact value is correct.", "criteria": {"answer_1": "20", "answer_3": "42", "answer_2": "14", "answer_0": "56"}}}, "targets": {"q0": [0.0, 1.0, 0.0, 0.0]}, "family": "jb_multihop"}